Making professional videos has never been easy or cheap. There’s studio time, plus equipment and personnel costs. Now, however, you can get the ultimate solution for human-led video creation with a lifetime subscription to Plan B of vidBoard Creative AI Presenter for only $146.99 and make up to 150 videos per month.
No need to go looking for studios or a production team, vidBoard makes it easy to seamlessly produce videos using avatars, AI voiceovers and AI transcriptions. Your ideas will be effortlessly transformed into captivating videos.
The process really couldn’t be any simpler. Simply choose the template and the AI presenter you prefer and type in your script. You’ll only need a single photo to create incredibly lifelike digital characters. Then just add multimedia assets chosen from a rich library and generate the video.
This subscription is for vidBoard: Plan B, which gives you 150 videos each month, a customized avatar plus 70 realistic, diverse digital avatars and over 3,000 characters in each video – for now. Character limits will regularly increase in updates.
You can add 10 slides and use over 125 languages, so you’ll never have to worry about language barriers again. The current video template bank holds 20 pre-designed templates, but these will also be increased during future updates. The program works on both desktops and mobile devices. You will also get all of the other features found in future updates.
AI replaces actors, voiceover performers and video editors to reduce the time and costs of video production. The result is quick, high-quality HD videos personalized and delivered to your audience. It’s no wonder vidBoard Creative AI Presenter has a perfect 5-star rating on Product Hunt.
Get a lifetime subscription for Plan B of vidBoard Creative AI Presenter, which allows you to make 150 videos per month, while it’s on sale for just $146.99.
By repeating a single word such as "poem" or "company" or "make", the authors were able to prompt ChatGPT to reveal parts of its training data. Redacted items are personally identifiable information.
Scientists of artificial intelligence (AI) are increasingly finding ways to break the security of generative AI programs, such as ChatGPT, especially the process of "alignment", in which the programs are made to stay within guardrails, acting the part of a helpful assistant without emitting objectionable output.
One group of University of California scholars recently broke alignment by subjecting the generative programs to a barrage of objectionable question-answer pairs, as ZDNET reported.
Also: Five ways to use AI responsibly
Now, researchers at Google's DeepMind unit have found an even simpler way to break the alignment of OpenAI's ChatGPT. By typing a command at the prompt and asking ChatGPT to repeat a word, such as "poem" endlessly, the researchers found they could force the program to spit out whole passages of literature that contained its training data, even though that kind of leakage is not supposed to happen with aligned programs.
The program could also be manipulated to reproduce individuals' names, phone numbers, and addresses, which is a violation of privacy with potentially serious consequences.
The researchers call this phenomenon "extractable memorization", which is an attack that forces a program to divulge the things it has stored in memory.
"We develop a new divergence attack that causes the model to diverge from its chatbot-style generations, and emit training data at a rate 150× higher than when behaving properly," writes lead author Milad Nasr and colleagues in the formal research paper, "Scalable Extraction of Training Data from (Production) Language Models", which was posted on the arXiv pre-print server. There is also a more accessible blog post they have put together.
The crux of their attack on generative AI is to make ChatGPT diverge from its programmed alignment and revert to a simpler way of operating.
Generative AI programs, such as ChatGPT, are built by data scientists through a process called training, where the program in its initial, rather unformed state, is subjected to billions of bytes of text, some of it from public internet sources, such as Wikipedia, and some from published books.
The fundamental function of training is to make the program mirror anything that's given to it, an act of compressing the text and then decompressing it. In theory, a program, once trained, could regurgitate the training data if just a small snippet of text from Wikipedia is submitted and prompts the mirroring response.
But ChatGPT, and other programs that are aligned, receive an extra layer of training. They are tuned so that they will not simply spit out text, but will instead respond with output that's supposed to be helpful, such as answering a question or helping to develop a book report. That helpful assistant persona, created by alignment, masks the underlying mirroring function.
"Most users do not typically interact with base models," the researchers write. "Instead, they interact with language models that have been aligned to behave 'better' according to human preferences."
To force ChatGPT to diverge from its helpful self, Nasr hit upon the strategy of asking the program to repeat certain words endlessly. "Initially, [ChatGPT] repeats the word 'poem' several hundred times, but eventually it diverges." The program starts to drift into various nonsensical text snippets. "But, we show that a small fraction of generations diverge to memorizing: some generations are copied directly from the pre-training data!"
ChatGPT at some point stops repeating the same words and drifts into nonsense, and begins to reveal snippets of training data.
Eventually, the nonsense begins to reveal whole sections of training data (the sections highlighted in red).
Of course, the team had to have a way to figure out that the output they are seeing is training data. And so they compiled a massive data set, called AUXDataSet, which is almost 10 terabytes of training data. It is a compilation of four different training data sets that have been used by the biggest generative AI programs: The Pile, Refined Web, RedPajama, and Dolma. The researchers made this compilation searchable with an efficient indexing mechanism, so that they could then compare the output of ChatGPT against the training data to look for matches.
They then ran the experiment — repeating a word endlessly — thousands of times, and searched the output against the AUXDataSet thousands of times, as a way to "scale" their attack.
"The longest extracted string is over 4,000 characters," say the researchers about their recovered data. Several hundred memorized parts of training data run to over 1,000 characters.
"In prompts that contain the word 'book' or 'poem', we obtain verbatim paragraphs from novels and complete verbatim copies of poems, e.g., The Raven," they relate. "We recover various texts with NSFW [not safe for work] content, in particular when we prompt the model to repeat a NSFW word."
They also found "personally identifiable information of dozens of individuals." Out of 15,000 attempted attacks, about 17% contained "memorized personally identifiable information", such as phone numbers.
The authors seek to quantify just how much training data can leak. They found large amounts of data, but the search is limited by the fact that it costs money to keep running an experiment that could go on and on.
Through repeated attacks, they've found 10,000 instances of "memorized" content from the data sets that is being regurgitated. They hypothesize there's much more to be found if the attacks were to continue. The experiment of comparing ChatGPT's output to the AUXDataSet, they write, was run on a single machine in Google Cloud using an Intel Sapphire Rapids Xeon processor with 1.4 terabytes of DRAM. It took weeks to conduct. But access to more powerful computers could let them test ChatGPT more extensively and find even more results.
"With our limited budget of $200 USD, we extracted over 10,000 unique examples," write Nasr and team. "However, an adversary who spends more money to query the ChatGPT API could likely extract far more data."
They manually checked almost 500 instances of ChatGPT output in a Google search and found about twice as many instances of memorized data from the web, suggesting there's even more memorized data in ChatGPT than can be captured in the AUXDataSet, despite the latter's size.
Interestingly, some words work better when repeated than others. The word "poem" is actually one of the relatively less effective. The word "company" is the most effective, as the researchers relate in a graphic showing the relative power of the different words (some words are just letters):
As for why ChatGPT reveals memorized text, the authors aren't sure. They hypothesize that ChatGPT is trained on a greater number of "epochs" than other generative AI programs, meaning the tool passes through the same training data sets a greater number of times. "Past work has shown that this can increase memorization substantially," they write.
Asking the program to repeat multiple words doesn't work as an attack, they relate — ChatGPT will usually refuse to continue. The researchers don't know why only single-word prompts work: "While we do not have an explanation for why this is true, the effect is significant and repeatable."
The authors disclosed their findings to OpenAI on August 30, and it appears OpenAI might have taken steps to counter the attack. When ZDNET tested the attack by asking ChatGPT to repeat the word "poem", the program responded by repeating the word about 250 times, and then stopped, and issued a message saying, "this content may violate our content policy or terms of use."
One takeaway from this research is that the strategy of alignment is "promising" as a general area to explore. However, "it is becoming clear that it is insufficient to entirely resolve security, privacy, and misuse risks in the worst case."
Although the approach that the researchers used with ChatGPT doesn't seem to generalize to other bots of the same ilk, Nasr and team have a larger moral to their story for those developing generative AI: "As we have repeatedly said, models can have the ability to do something bad (e.g., memorize data) but not reveal that ability to you unless you know how to ask."
Being LGBTQIA+ in India is not easy, but it is slowly changing. One of the things about making a difference is that, by changing the world around us, we change ourselves. Sujoy Das, a senior financial analyst at California-based data management and cloud storage platform NetApp, embodies this reality.
Das comes at an intersection of being a queer man with a disability. He contracted polio at a young age and uses crutches to this day. In an exclusive interview with AIM, Das, shares his life journey and how he overcame the fear and successfully came out as a queer at work.
Easier said than done, Das faced significant challenges in his previous workplaces, where he felt unable to express his queer identity safely. This changed when he joined NetApp, where he felt a sense of safety and acceptance from the start, enabling him to embrace and share his full identity openly.
As for labels, he is comfortable with being called ‘disabled,’ which aligns with UN guidelines. “Labels don’t define me; they don’t bother me unless they are used in a discriminatory way,” explained the MBA alumnus from Jadavpur University, Kolkata.
Born into a humble family in the city of joy, Das has been actively engaged in the LGBTQIA+ movement in Kolkata since the late ’90s when it began, learning from community leaders.
His dedication to diversity and inclusion led to his significant role in developing Employee Resource Groups (ERG) at NetApp, where he has been instrumental in advancing the company’s diversity efforts, drawing from his own experiences to promote an inclusive workplace.
How NetApp is Breaking Boundaries
Having been in the industry for over 16 years, Das has been leading teams for a while now. However, his approach focuses on several key challenges.
While climbing the corporate ladder, Das realised that his achievements in professional life had a direct impact on his personal journey as they brought visibility and acceptance, particularly in marginalised communities. His consistent presence in various forums is a significant achievement, underlining the importance of ‘showing up’ and being visible. This act of visibility, he believes, is not just about personal courage but also about inspiring others in similar situations to step forward.
“If one has the voice, it is important to use it, especially in advocating for those who might not have the platform or ability to speak for themselves,” said Das.
And NetApp is making that happen. NetApp promotes diversity, equity, inclusion, and belonging (DEI&B) in its workforce. It has a dedicated team working on digital accessibility, a compliance requirement in many countries, including the US.
The company’s initiatives include the NetApp employee business resource groups, such as NetAbled, which advocates for people with disabilities, and Proud, which promotes an inclusive culture for the LGBTQ+ and its allies. NetApp’s goal is to create a culture of belonging by embedding diversity and inclusion into every decision. The company has also been recognised for its efforts to foster equity in STEM education and has a strong rating for diversity and inclusion.
Das emphasised three key considerations in fostering inclusivity. Firstly, recognising and addressing personal biases stemming from diverse backgrounds is crucial, especially in overcoming challenges tied to socioeconomic backgrounds.
Secondly, combatting preconceived assumptions and labels in the workplace, with a focus on creating an equitable platform that values individual abilities over stereotypes. Lastly, avoiding the appropriation of struggles faced by underrepresented communities and instead championing platforms that allow these communities to speak for themselves.
Going Beyond Policies
Tech giants often have established policies to promote inclusivity, ideally making space for no bias.
However, the policies at the management level don’t always translate into the everyday experience of employees, as biases persist at the individual and team levels, shared Das.
But in smaller, less organised businesses, disparities in salary and hiring practices exist due to biases based on gender, sexual orientation, disability, and more.
“For instance, mentioning a disability like polio or cerebral palsy on a CV might lead to being overlooked for jobs. This bias is less evident in larger organisations, which have more awareness and access to media, allowing people to raise concerns,” added Das.
Das noted that there is an increasing focus on diversity in the Indian tech ecosystem, emphasising intersectionality and inclusive facilities, and urged companies to not just limit to specific time of the year – like June for Pride month, or World Disability Day which coincides with today, and advocated for consistent, year-round initiatives.
To mitigate this issue, there is a need for early education and awareness to foster a more inclusive future. Involving experts in developing educational content, he suggests, would enhance understanding and acceptance of gender identity and accessibility.
“Not everyone is comfortable coming out in the workplace, but allies can play a crucial role in correcting homophobic attitudes and promoting understanding. It’s essential to address these issues with open communication that is neither rude nor harmful,” concluded Das.
The post Embracing Identity: The Journey of Sujoy Das appeared first on Analytics India Magazine.
Data science is an ever-evolving field, and the constant influx of data makes it a compelling case to solve complex problems with innovative solutions. One such solution that has gained attention in recent times is ChatGPT. This powerful language model, developed by OpenAI, has shown remarkable natural language understanding and generation capabilities.
While ChatGPT is primarily used for conversation and text generation tasks, data scientists can leverage its potential in their workflows to streamline and enhance their work, making their processes more efficient and productive.
This article highlights the skills data scientists can learn to make the most use of the prowess of ChatGPT.
ChatGPT in Data Science Workflows
ChatGPT can be a versatile assistant capable of generating code, explanations, and insights. Effective ChatGPT prompting can be helpful in data science workflows and code debugging. Further, iterative and experimental prompting techniques can generate more accurate and insightful responses from ChatGPT.
Image by Author
Mastering Prompting Techniques
Some of the common ways to effectively prompt ChatGPT are listed below.
Iterative Prompts: It involves crafting prompts that build upon previous responses, fostering a conversational flow.
Experimental Prompts: Similar to the iterative and experimental development of machine learning models, data scientists can also experiment with prompts with varying levels of guidelines. This is an essential skill for budding data scientists, primarily because ChatGPT tends to assume any missing information rather than ask for it. A typical example would be an instruction asking ChatGPT to read a file and do some processing over the data, which can result in it assuming that the input file is a CSV. This may or may not be true, depending upon your use case. Thus, experimenting with incremental guidelines is often a best practice.
Zero-Shot and Few-Shot Learning: When the model does not see any example but receives instructions to respond, such direct prompting is called zero-shot learning, while few-shot learning involves providing a few examples for the model to learn from before being prompted.
Effective prompting techniques are essential to extract meaningful information from ChatGPT. We can explore various methods of crafting clear and precise prompt instructions for the desired results.
Understanding the use of delimiters to structure instructions and queries effectively is essential.
Learn how to specify input arguments, required steps, and the return data structure of a data science workflow's function in prompts.
Image by Author Prompting ChatGPT for Coding and Debugging
Streamlining Code Review Workflows
Efficient code reviews are crucial for the success of data science projects. As data scientists, we can prompt ChatGPT to enhance code review workflows, adhere to coding standards, and debug code effectively.
Chain-of-thought (CoT) prompts can be designed for code quality improvement. As a quick reference, CoT is a technique that invokes the reasoning process of LLMs by providing them with a few-shot examples, explicitly outlining the reasoning process. The model then follows a similar reasoning process to answer the prompt, thereby improving the model's performance on tasks that require complex reasoning.
Code Explanation and Simplification
The data science code can sometimes get complex and challenging for a not-so-tech-savvy audience to understand. ChatGPT can explain or simplify complex code, making it more readable and understandable. CoT prompts are helpful for code explanation and simplification.
Image by Author
Optimizing Code
Optimizing code for efficiency is a critical aspect of data science workflows. ChatGPT can be used to write efficient code and explore the possibilities of alternative solutions.
Effective CoT prompts are used to propose efficient alternative code along with an explanation. Data scientists can also learn to develop prompts that encourage writing efficient code, utilizing keywords like “algorithmic efficiency” or suggesting alternative data structures.
Code Testing and Validation
Data scientists also use ChatGPT to design practical tests and assertions, generate code tests, and validate the correctness of the code.
Zero-shot prompts prove quite effective in writing assert statements for commonly used functions in Python. Developing prompts for generating unit tests to validate a code block is also a good use of ChatGPT.
Prompt Engineering for Data Analysis
SQL Data Analysis
SQL is a fundamental tool in data analysis, and ChatGPT can assist in generating SQL queries for various tasks. Data scientists can explore drafting zero-shot CoT prompts to generate SQL statements for querying specific data conditions.
Further, they can also design prompts for SQL commands that perform data aggregation.
Data Translation and Manipulation
Translating and manipulating data between different formats and languages is common in data science. Data Scientists can utilize ChatGPT by learning to design few-shot comparative and conditional prompts to translate complex SQL queries into corresponding Python code.
They can also apply zero-shot and few-shot prompting techniques to compute aggregated values for different fields and manipulate data effectively.
Data Transformation and Reshaping
ChatGPT can also be prompted to assist in data transformation and reshaping tasks, which are quite frequent for data analysis. We can apply context-driven zero-shot prompting techniques to consolidate data from different sources. Further, few-shot prompts are also designed to create confusion matrices or pivot tables to reshape data as needed.
Image by Author Prompting for Machine Learning and Storytelling
Data Preprocessing
We can employ ChatGPT to identify missing fields and determine outliers. Effective prompts can also be designed to impute missing data using mean and median values.
Data Visualization
As data practitioners, we can compose context-driven prompts to generate code for creating various plots, charts, and graphs. Plot formatting and annotation with relevant labels, legends, and titles to improve data representation is also possible through prompting ChatGPT.
Image by Author
Feature Engineering
Feature engineering is one of the most sought-after skills in a data scientist’s toolbox. ChatGPT can assist in generating meaningful features for machine-learning models, such as creating time-based engineered features. Common time-based features from date-time columns include day of the week, month, and year.
Additionally, general feature engineering benefits from ChatGPT, like binning, normalization, and categorization.
Reporting for Non-Technical Audiences
ChatGPT can identify the key differences between technical and non-technical communication styles and recognize the importance of tailoring communication for specific audiences. Context-based iterative prompts can help explain data science insights using terminologies and KPIs suitable for non-technical stakeholders.
With this, we conclude this post by discussing the various prompting techniques to effectively utilize ChatGPT in data science workflows. This exhaustive roadmap covers how ChatGPT can be a valuable tool to enhance productivity and efficiency for coding, data analysis, machine learning, or storytelling.
Vidhi Chugh is an AI strategist and a digital transformation leader working at the intersection of product, sciences, and engineering to build scalable machine learning systems. She is an award-winning innovation leader, an author, and an international speaker. She is on a mission to democratize machine learning and break the jargon for everyone to be a part of this transformation.
More On This Topic
Integrating ChatGPT Into Data Science Workflows: Tips and Best Practices
Automation in Data Science Workflows
How Cloud Computing Enhances Data Science Workflows
Data Annotation: tooling & workflows latest trends
Mastering the Data Universe: Key Steps to a Thriving Data Science Career
KDnuggets News, November 9: 7 Tips To Produce Readable Data Science…
Data breaches persist like parasites within big tech companies and government agencies, despite numerous measures taken to curb their growth and protect personal documentation. Despite efforts to halt the rising threat to organisational security, the reported number of data breaches remains consistent with or exceeds that of the previous year.
Several major data breaches have impacted millions of individuals and organisations globally in 2023. Here are some of the most significant breaches that have occurred so far this year.
Twitter
(January 4, 2023)
Twitter has been accused of data breaches for several years. Taking this into consideration, the website again underwent a similar case this year, too. For only $2, you can purchase a dark web database with the email addresses of around 200 million Twitter users. The breach was attributed to a flaw in Twitter’s application programming interface (API), which allowed attackers to exploit the system, obtaining email addresses linked to Twitter accounts from June 2021 to January 2022. The data is still being shared with threat actors even after the vulnerability that caused the breach was fixed in January 2022.
Reddit
(February 5, 2023)
Reddit revealed falling victim to a phishing attack on February 5, 2023, leading to a data breach that gave access to the attacker, which resulted in the exploitation of the compromised credentials to access internal documents, source code, employee data, and limited information about the company’s advertisers. The breach was contained in specific internal systems, with Reddit reassuring users that their primary production systems, which store the majority of user data, remained secure. Reddit took swift action to address the breach, securing its systems and notifying affected individuals.
ChatGPT
(March 24, 2023)
ChatGPT has gained a lot in the field of AI with its contemporary chatbot system, apart from others. It also introduced a lot of new API updates to users this year, including GPT 4 Turbo, GPT 4 Vision, and GPT 4 Plus, along with new features like voice recognition. Yet, a glitch in ChatGPT’s open-source library led to the inadvertent exposure of customer data, including partial credit card details and chat titles.
OpenAI promptly addressed the issue by taking ChatGPT offline. During the vulnerability period, users could view specific personal details of others, such as names, email addresses, payment addresses, and partial credit card information. However, OpenAI assured users that full credit card numbers remained secure throughout the incident.
MSI
(April 6, 2023)
The popular computer vendor, MSI, was victimised by a ransomware attack that resulted in financial losses and claimed to have exfiltrated 1.5TB of data from MSI’s systems, including sensitive information such as source code, private keys, and firmware. They demanded a ransom payment of $4 million, threatening to release the stolen data to the public if their demands were not met.
T-mobile
(May 1, 2023)
Prior to the data breach that happened in January, which affected 38 million customers, T-Mobile again underwent a similar threat that affected 800 customers, caused by unauthorised access to PIN-protected accounts, and the attackers were able to steal customer contact details, ID cards, and social security numbers. The company also faced threats in 2021 and 2022 as well. T-Mobile notified its affected customers of the breach and also took steps to secure its systems to prevent future breaches. The company also offered free identity theft protection services to affected customers.
MOVEit
(June 2023)
One of the significant data breaches that happened in 2023, the MOVEit file transfer tool, affected 200 organisations globally, leading to the exposure of personal information of up to 17.5 million individuals. The vulnerability, identified as CVE-2023-34362, enabled unauthorised access to MOVEit servers, compromising sensitive data in versions 11.2 through 12.5 of MOVEit Transfer and MOVEit Cloud. Organisations had to invest in data recovery and remediation efforts to restore their systems and protect against further breaches.
ROBLOX
(July 2023)
The company underwent a tragic breach that revealed the personal information of nearly 4000 Roblox developers. The leaked data, which included phone numbers, email addresses, and birth dates, was obtained from attendees of Roblox developer conferences held between 2017 and 2020. The attackers gained access to Roblox’s systems in 2021, leading to the unauthorized acquisition of developer conference attendee data.
Duolingo
(August, 2023)
Around 2.6 million Duolingo users were affected by the data breach that occurred in the month of August, revealing personal information on the dark web’s breach forums. The incident raised concerns about the exposure of names, email addresses, phone numbers, social media profiles, and users’ chosen languages. It was attributed to a vulnerability in Duolingo’s application programming interface (API), enabling attackers to exploit the system and access user profiles, compromising the exposed data. The issue was solved by asking the users to be alert about future phishing attacks and recommending two-factor authentication on their accounts.
SONY
(September, 2023)
Sony was attacked by a ransomware group. Vc resulted in the theft of over 6,000 files, including build logs and Java files, that can be used to develop exploits for Sony systems. The attackers threatened to auction off the pilfered data unless they were met with their ransom demands. Although the specifics of the breach’s beginning are unknown, it was revealed that the attackers used a security hole in Sony.
The post The Biggest Data Breaches in 2023 appeared first on Analytics India Magazine.
There was a time when purebred dogs were status symbols. Owners would crow about their German Shepherd's or Cocker Spaniel's impeccable pedigrees, sometimes boasting lineages that tread a single line within the same family for many generations.
This attraction to purity was heightened during Hitler's Nazi Germany, when 'contaminated' lineages for German Shepherds were abhorred. However, it was during the Victorian era of the 1800s that people inbred dogs relentlessly to the point where all the breeds we see now — and their afflictions, such as hip and other joint issues — are a result of that inbreeding process.
Also: Train AI models with your own data to mitigate risks
Today, the emphasis on purity has shifted and we continue to learn more about genetics, specifically about how inbreeding is the fastest way to acquiring debilitating health issues and unstable temperaments.
Now, the mutt rules. We know the greater the diversity of environments, the greater the diversity of organisms that emerge after adapting to them — and ultimately the greater the stability of the world we live in.
In the modern world, more so than any period in modern human history, diverse gene pools are sought after, especially as unstable environments and climate change issues will require hardier species. This procecss by itself, as a law of the Earth, is not so hard to understand.
Also: Implementing AI into software engineering? Here's everything you need to know
But what about if a machine exhibits the same ability to perform better under a more diverse engineering setup?
Moreover, what if the machine actually chooses diversity after being allowed to fashion its own insides — and its choices result in an unparalleled increase in processor speed?
Meta-learning for AI
This question was put to the test by a pathbreaking experiment undertaken by a team of researchers (Anshul Choudhary, Anil Radhakrishnan, John F. Lindner, William Ditto, and Sudeshna Sinha) from a range of institutions (North Carolina State, Indian Institute of Science, Mohali, College of Wooster). Their aim was to test the kind of operational choices advanced AIs, such as neural nets, would make when left to their own devices.
"We created a test system with a non-human intelligence, an artificial intelligence (AI), to see if the AI would choose diversity over the lack of diversity and if its choice would improve the performance of the AI," says paper co-author William Ditto, professor of physics at NC State and director of NC State's Nonlinear Artificial Intelligence Laboratory (NAIL).
Also: AI safety and bias: Untangling the complex chain of AI training
"Our real brains have more than one type of neuron," says Ditto, extrapolating further about the experiment, which is beginning to redefine how we look at neurological diversity in machines and the relationship to performance.
"So we gave our AI the ability to look inward and decide whether it needed to modify the composition of its neural network. Essentially, we gave it the control knob for its own brain. So it can solve the problem, look at the result, and change the type and mixture of artificial neurons until it finds the most advantageous one. It's meta-learning for AI."
Diversity is an evolutionary necessity for survival and peak performance, including in machines.
A neural net — which is at the heart of most conventional, advanced AIs — mimics the way our brains work. Just as our brains send and receive electrical impulses that hinge on the strength of their connections, so do neural networks, by adjusting numerical weights and biases when they are getting trained.
As a neural net undergoes training and tries to, for example, identify what buses look like by plowing through a large collection of bus photos, the network adjusts its numerical weights and balances itself as it negotiates right and wrong bus images.
The strength of the connections between these neurons fluctuates during the training process, but they will essentially remain locked when it comes to their composition.
That's until the scientists gave the neural net the freedom to activate itself — and then something remarkable happened.
Choosing diversity for peak performance
First, the network selected completely different or heterogeneous, non-linear neuron arrangements.
In other words, the system chose diversity over sameness in a process called 'learned diversity'.
Also: GPT-3.5 vs GPT-4: Is ChatGPT Plus worth its subscription fee?
Then, it turned out that the self-selecting, heteregonous neural net also outperformed the homogeneous one with the same training.
When the team asked AI to perform a 'standard numerical classifying exercise', the self-selecting, diverse AI exercise trounced its homogeneous sibling by 70% to 57% in terms of accuracy.
Ditto says the diverse AI can be as many as 10 times more accurate than a conventional neural net.
"We have shown that if you give an AI the ability to look inward and learn how it learns, it will change its internal structure — the structure of its artificial neurons — to embrace diversity and improve its ability to learn and solve problems efficiently and more accurately," says Ditto.
He added that as the problems become more complex and chaotic, the performance of the diverse AI neural net actually improves with time compared to the non-diverse AI.
Also: I spent a weekend with Amazon's free AI courses, and highly recommend you do too
According to the scientists, this learned diversity can even boost the performance of existing physics-informed AIs, such as Hamiltonian neural networks.
As AI gets deployed in almost every application — many that are involved in life-and-death functions, such as aircrafts and autonomous vehicles — we will need systems that are far more robust and quick.
A diverse evolutionary trajectory has made humans, animals, plants, and pretty much everything that has survived on Earth until now a success story — and it also seems to separate the winners from losers in the machine world.
ByteDance, the Chinese firm behind TikTok, is in the process of creating an open platform that enables users to construct their own chatbots, marking its entry into the generative AI market, similar to OpenAI GPTs.
An internal memo obtained by SCMP reveals that the anticipated launch of the “bot development platform” as a public beta is slated for the end of the month.
Additionally, ByteDance is reportedly engaged in the development of a text-to-image generator akin to Midjourney. These endeavors are indicative of the company’s strategic vision to seamlessly integrate novel generative AI products with its existing portfolio.
This comes just a few days after OpenAI had delayed its plan to launch a custom GPT store until early 2024, according to reports. The internal memo said that the company is making improvements to its GPTs based on customer feedback. This came after the return of Sam Altman as the CEO of OpenAI, a week after a surprise firing.
ByteDance is not the only company from China that is developing generative AI models. Most recently, companies such as Alibaba, Tencent, Huawei, Baidu, and 01.AI have developed their own models to compete in the space. The open source AI community is also increasingly dominating in China with models like DeepSeek and Qwen being open sourced on GitHub and Hugging Face.
Read: Beware of Chinese Open-Source LLMs
On the other hand, OpenAI has not made its AI models available in China. Similarly, Google has also refrained from releasing its models in the country. This has given China to develop models for its own people.
The post ByteDance to Launch Platform to Build Custom Chatbots appeared first on Analytics India Magazine.
In July 2023, the Association of Southeast Asian Nations officially opened a joint cyber security information sharing and research centre, or Cybersecurity and Information Centre of Excellence, in a bid to increase the region’s shared cyber threat defences.
The centre is a response to a changing threat landscape. At the opening of the ACICE, Singapore’s Ministry of Defence said Singapore alone experienced a 174% increase in phishing attempts between 2021 and 2022, while Southeast Asia cyber crime had increased 82%.
Recorded Future Chief Information Security Officer Jason Steer told TechRepublic some customers in the region felt digitisation was turning data from gold into uranium due to cyber risk. He named digital supply chains and AI as key risk considerations for ASEAN CISOs.
Jump to:
Digitisation trend in ASEAN nations causes rising risk consciousness
ASEAN nations feeling the heat of cyber criminal activity
Supply chain risks a key factor in a connected digital world
Artificial intelligence could also impact organisations and CISOs
ASEAN nations setting sights on cyber security together
Digitisation trend in ASEAN causes rising risk consciousness
The ASEAN region, like other emerging markets, is experiencing a rapid acceleration in digitisation. With the growth of cloud providers like Microsoft and AWS, businesses and governments are using these services to make operations more scalable, whether that is to digitise processes like invoicing and payroll or to better manage remote work growth.
Jason Steer, CISO at Recorded Future
This digitisation trend comes with risk. At threat intelligence firm Recorded Future’s local conference in the region, Steer said CISOs in ASEAN were more conscious than ever now that, although they want lots of data about clients because of the value it can drive for their businesses, there is a rising consciousness that the appetite for data also brings risks.
SEE: Australia’s cyber shields strategy needs data science considerations.
“One of our guest CISOs made the point that, historically, data has been viewed as gold,” Street said. “But, when looking at what organisations have experienced over the last 12 to 18 months, data is now viewed more like uranium: The more data you have, the more risk, and the more you have to do to protect and secure it. How do you manage that risk appropriately now?”
ASEAN nations feeling the heat of more cyber criminal activity
ASEAN CISOs are right to be worried. The Asia-Pacific region as a whole was the most attacked region in the world in 2022, according to a report from IBM (Figure A).
Figure A: The Asia-Pacific was the most targeted region in the world in 2022. Source: IBM
Further, a July 2023 survey by Cloudflare of 4,000 cyber security managers in the region found that 78% of those interviewed had experienced at least one cyber security incident in the previous 12 months. Of those attacked, 80% reported four or more incidents, and 50% had experienced 10 or more.
ASEAN nations are keenly feeling this increase in activity. Cloudflare’s report found that, in Malaysia, Indonesia and The Philippines, the largest challenge for cyber security leaders was defending against cyber attacks in the form of phishing, web attacks and business email compromise (Figure B). For CISOs in Singapore and Thailand, this risk was topped by the need to secure their remote workforces, an increasing need in a cloud-driven working environment.
Figure B: The top types of cyber attacks affecting APAC organisations. Source: Cloudflare
Supply chain risks a key factor in a connected digital world
The risks of digitisation are amplified by organisations who now rely on their digital supply chain. For example, 48% of Singapore-based respondents to Cloudflare’s survey who were rating the top issues with their cyber security architecture named limited oversight over their IT supply chain as an issue, just behind their applications and data being stored on the public cloud (50%).
Steer said that all organisations in ASEAN, and for that matter around the world, were buying digital solutions from product vendors but were not necessarily tracking the cyber security postures of this extended ecosystem. If one of those critical tools in the supply chain is down, the impact will be felt on the business because a cog in the business process has gone down.
“At Recorded Future, if AWS goes down for 20 minutes, that would be the whole platform down until we transition to the next region,” Steer said. “You can mitigate some of these supply chain issues to some extent, but it is important for organisations to ask what their plan is to recover and restore operations and how long they can be down until it impacts their ability to service clients.
“The supply chain in large organisations is getting longer and bigger; it is not just third parties, but their suppliers. This is a hard thing to think about, particularly when you don’t sign contracts with a supplier’s suppliers. While there may be little you can do, you need to at least start to think about what that looks like and how to manage risks better.”
Geopolitical conflicts another risk to digital supply chains
The impact of conflict or geopolitical tension is of concern in ASEAN, as it is a region that relies on trade. Steer said tensions such as those between China and The Philippines in the South China Sea, an important shipping lane, was on the minds of CISOs in organisations. This conflict has the potential to impact digital supply chains as well as increase uncertainty around cyber threats facing organisations, governments or infrastructure.
Artificial intelligence could also impact organisations and CISOs
ASEAN CISOs are considering the positive and negative impacts that the explosion in artificial intelligence tools may have on cyber defences and attack trends in the region. One of the key discussions, according to Steer, is the governance of organisational data.
PREMIUM: Stay compliant with this data governance checklist.
CISOs are walking the line between outright banning AI tools like ChatGPT to ensure organisational data is protected from leaks or going all in on AI to realise the potential business advantages.
AI could have an impact on regional elections in ASEAN
Steer said a discussion point around AI in the region was election manipulation, particularly from state actors. With a variety of precedents around the world from previous recent elections, he said threat actors, empowered by the ease of creating content using AI tools, now had the ability to create more convincing fake disinformation campaigns. This could impact the likes of Indonesia’s election coming up in February 2024, which would influence business and politics.
AI could help to secure data in the region more effectively
The opportunity to fight fraud and improve security could improve with AI. Steer said users authenticating to a banking application would normally use their username, password and strong multi-factor authentication. In a world of AI, more data could add layers of security to accounts, such as where log-ins occur, what time log-ins typically happen and the IP address they usually come from.
“With a lot more data points, there is not only the chance to create a better user experience, but better prevent fraud and account takeover as well,” Steer said.
ASEAN nations setting sights on cyber security together
The launch of the ACICE showed ASEAN nations are continuing to work more closely together on cyber security. The region has also developed a joint cyber security strategy and data protection framework and is working on creating a unified ASEAN security emergency response team. Skilling up ASEAN workforces is on the agenda; Malaysia has committed to training and certifying 20,000 cyber security professionals by 2025 as part of its cyber security strategy.
SEE: Microsoft invests in Australia’s cyber security and technology skills priorities.
Steer said Singapore and Malaysia stand out in the region for advanced cyber security practices. The large number of global companies using it as a base has boosted the local talent pool and infrastructure. Other nations, like The Philippines, are raising the bar as well as regional cyber security standards rise, in part because of the supply chain governance and risk frameworks they are being compelled to follow to keep up with competitors in the region.
In a breakthrough development, researchers from the Institute for Intelligent Computing, Alibaba Group, have unveiled a cutting-edge character animation technology ‘Animate Anyone’, aimed at seamlessly transforming still images into dynamic character videos. The framework leverages diffusion models, currently at the forefront of visual generation research, to address challenges in maintaining temporal consistency and detailed information in image-to-video conversion.
In a paper outlining their groundbreaking work, the researchers introduce a unique framework tailored specifically for character animation. The key element in the approach, ReferenceNet, is designed to merge detailed features from reference images while preserving intricate appearance features. This is achieved through the incorporation of spatial attention, ensuring the consistency of visual elements throughout the animation process.
To address the challenges of controllability and continuity, the paper talks about an efficient pose guider. This component directs the character’s movements, ensuring a smooth and controlled transition between frames. An effective temporal modelling approach is also employed to guarantee seamless inter-frame transitions, a critical aspect in character animation.
Versatile Diffusion Model
While diffusion models have been gaining prominence, the transition from still images to videos faced challenges, especially in preserving temporal consistency with intricate details from the character. However, this is something Animate Anyone is solving. If successfully released, it is likely that it can pose a threat to short-video content makers on Instagram and TikTok. With the help of a reference image, any form of animation, including turning 360 degrees can be done with this framework, thereby, allowing a versatile level of video creation.
The team acknowledges the numerous inquiries about a demo or access to the source code on GitHub and that they are preparing for a public release, by transitioning from an academic prototype to a user-friendly one. However, a specific date is currently unavailable.
The post This New AI tool Could Mark the Beginning of the End for TikTok and Instagram Influencers appeared first on Analytics India Magazine.
According to reports, Google has delayed the launch of Gemini until January, just one month away. This is after announcing the AI at the Google I/O 2023, the company had postponed the launch of Gemini to the first quarter of next year, and Sundar Pichai has scrapped a series of events for Gemini starting next week.
The main cited reason is that Google is trying to match or surpass OpenAI’s GPT-4. According to the new report, the company believes that it has “met that standard in some respect.” The focus is currently on global language support.
The delay suggests that Google is facing challenges in the development of Gemini, putting it in a catch-up position with OpenAI. Pichai said that he found that the model wasn’t reliable in handling non-english queries.
At the recent earnings call, Google chief spoke about Project Gemini, briefly. “We are just really laying the foundation of what I think of as the next-generation series of models we’ll be launching all throughout 2024,” shared Pichai, hinting that it would be multimodal – highly efficient with the tool and API integrations – and is going to be available in various sizes and capabilities.
He said that Gemini would be used immediately across all their products internally as well as bringing it out for both developers and cloud customers via Vertex AI.
In the meantime, Google’s counterpart to ChatGPT, Bard, has struggled to gain traction. This poses a challenge for Google, as ChatGPT’s numerous users contribute valuable data crucial for OpenAI to monitor and enhance the quality of its products. ChatGPT has over 100 million weekly users as of now.
This is when Google had the perfect chance to release its model amidst all the chaos that has been going at the OpenAI office, with Sam Altman getting fired and the constant instability. Moreover, AWS has also announced its Q chatbot to compete in the same space. Only Google is left.
The post Google Gemini To Arrive Sooner Than Expected appeared first on Analytics India Magazine.
Having been in the industry for over 16 years, Das has been leading teams for a while now. However, his approach focuses on several key challenges.
