Рубрика: AI

I was excited, of course.

Earlier this year, Microsoft invited me to its Silicon Valley campus to contemplate AI, the generative kind.

So there I was in this enormous, airy building of the future, with hardly a human being in sight. I wafted into a very fetching theater – yes, they spelled my name wrong on the big screen — and there were several actual journalistic luminaries on stage too.

Also: The future of generative AI: Here's what technology analysts are saying

Oh, and a PR man from Microsoft, dressed all in black.

The specific subject was AI and journalism. Little did I know how much it would move my whole year. For here I was, perhaps for the first time, confronted with the three great horsepersons of AI — the Experimenters, the Fearful and the Capitulists.

Those last ones? Capitalists who have already capitulated to AI's allegedly all-encompassing powers.

Let's try something new

Some panelists mused that the seemingly sudden incursion of AI offered a fine chance to dabble with it.

At one news service, for example, they've experimented with an AI version of one of its sports editors.

Also: ZDNET looks back on tech in 2023, and looks ahead to 2024

The idea is to put some text into this AI-generated expert and let it/them speak. Yes, but could this bot give us a decent betting line on the Niners vs the Cowboys? You have to find a balance, said the news service's journalist.

But is there really any value added by being, well, so tech-clever?

This seemed to me to be one of the currently less-considered questions about AI. The year was all about generative AI. But, as with quite a few tech breakthroughs, how useful (and profitable) will it ultimately be? And precisely where and how?

So a strong impulse was to put it to the test in as many ways as possible. You never know what might emerge.

Oh, AI. Who is the fairest of them all? You Are.

But then there was the gentlefellow from another tech publication.

I know that sounds like the beginning of a joke, but it truly isn't. It's more of a mood enhancer.

Also: The future of work is more human than you'd think, say these business experts

You see, this particular gentlefellow is an enormous enthusiast for AI. Actually, 'enthusiast' doesn't quite cover it. He seemed more like the mesmerized subject of a despotic kingdom.

He offered the prediction that "generative AI will touch every piece of every part of the process of news from ideation, to headline generation to story editing."

Full disclosure: I had the idea for this column myself, my editor is a human being with infinite patience, and I'm writing with my own fingers and (what's left of my) mind.

That's the thing with predictions. They don't always come true. Just ask any AI sports editor.

I won't dwell on some of this nice man's more vivid admissions — they've been covered elsewhere.

For me, what was most affecting were these words: "Frankly, I think that the AI model is always more clever than me because it includes all of the written text throughout all of history."

I fear several things shot through my head at this very moment. No, of course 'How clever actually are you?' wasn't one of them.

Also: The promise and peril of AI at work in 2024, according to Deloitte's Tech Trends report

Not even when he said: "It feels like it is so much more talented than I'll ever be."

I met many people in college who'd read a lot of books. Seemingly every book that mattered. But would I have thought them clever or talented? Would I have imagined they'd, um, change the world? Not really. (Update: They didn't.)

Let AI be AI but you do you

I confess that night became my tech moment of the year.

Everywhere I went I seemed to meet either experimenters, fearers or, indeed, capitulists.

Both experimentation and fear are, of course, understandable. But to witness capitulism live on stage was a touch startling, and the feeling hasn't quite gone away.

Also: AI in 2023: A year of breakthroughs that left no human thing unchanged

I couldn't help offering a small retort to this committed AI capitulist.

"Maybe you should just believe in yourself more," I said. "I'm really concerned about you. Is it possible that because of your enthusiasm, you're already abdicating your own talents? You're actually maybe better than you think."

It kept on striking me throughout the year that for all the talk about the evil side of AI – and, because humans, there's plenty of evil – there's another side.

The true fascination of generative AI doesn't necessarily lie in what it's going to do to us, or even for us. It's what we can do with it.

It may actually show how clever we really are, not merely how clever we think we are.

If the internet taught us anything – yes, I know the jury is still out – it's that we derived enormous benefits and created new ways of talking and being — just as we endured awful changes of mood, behavior and hope.

Isn't that likely with AI too?

Also: Generative AI filled us with wonder in 2023 — but all magic comes with a price

2023 wasn't necessarily the year when AI began to take over our souls. It was merely a large new door opening, sucking us toward a blinding light.

I feel an appropriate level of fear having experimented with ChatGPT, with both hilarious and freakish results. (Watching a deepfake version of someone you do business with is truly a mind-twisting experience.)

But to be a capitulist strikes me as not merely defeatist, but just a little dull.

Here's an oddly optimistic thought: AI, this clever and talented thing, might even help us slow down a little and think a little more.

Also: Here's how to create your own custom chatbots using ChatGPT

Oh, and here's another optimistic thought: We've just learned that AI can get tired and lazy – perhaps it's embracing the human condition in the same way that we're embracing the AI condition.

After all, what can it do without our input? (Please don't answer that right now.)

We won't know for sure, of course until, say, 2032.

You may, of course, place your bets now with my personal AI Betbot service.

Featured

The rise and rapid adoption of new innovative technologies, such as generative artificial intelligence, no-code apps, automation and the Internet of Things, have dramatically changed the global cybersecurity and compliance landscape for every industry.

Cybercriminals are turning to new techniques, tools and software to launch attacks and create greater damage. As a result, the 2023 Cybersecurity Ventures Cybercrime Report predicts a rapid increase in damage costs associated with cybercrime — projected to cost $10.5 trillion globally in damages by the end of 2024. The report lists cost of data breaches, stolen funds, intellectual property theft, operational disruptions and post-attack recovery as the main expenses for organizations under this trend.

On the other hand, Google’s Cloud Cybersecurity Forecast 2024 report highlights the increased use of AI to scale malicious operations, nation-state-supported cybercriminal gangs, zero-day vulnerabilities and modern phishing as main attack vectors for the coming year.

To stay ahead of the curve, IT and security leaders should focus on layered security solutions and zero trust to keep their companies’ data safe from top cybersecurity threats like ransomware and phishing.

Learn about Kolide

Jump to:

• 1. Ransomware
• 2. OT-IT security
• 3. Dark Web
• 4. Malware as a service and hackers-for-hire
• 5. Modern phishing
• 6. IoT and Industrial IoT
• 7. State-sponsored attacks
• Staying vigilant in the evolving threat landscape

1. Ransomware

Ransomware — the breaching of business-critical systems and assets with the goal of encrypting them and holding them for ransom — will continue to plague organizations across all sectors in 2024. New and established cybercriminal groups will leverage ransomware as a service, making it easier than ever to launch sophisticated attacks. They will also employ evolving extortion tactics like double and triple extortion, pressuring victims through data leaks.

SEE: Here’s everything you need to know about ransomware.

As proven by the November 2023 ransomware attack on MeridianLink by ALPHV/BlackCat ransomware group, ransomware gangs are also willing to manipulate regulations. In that attack, BlackCat reported its own crime to put pressure on MeridianLink leveraging the new U.S. Securities and Exchange Commission law.

Healthcare, government and critical infrastructure will be particularly targeted by ransomware. Organizations must prioritize ransomware defense by updating systems, implementing robust backups, training employees and considering cyber insurance. More importantly, companies must ensure their security teams and experts have all the resources they need and are not working under unsustainable pressure.

2. OT-IT security

The convergence of operational technology and information technology in critical infrastructures, industrial facilities, public service providers and manufacturing plants is creating new vulnerabilities and attack opportunities for cybercriminals. Attacks on OT infrastructures via IT-compromised systems can disrupt operations, cause physical damage and risk public safety.

Notable 2023 OT-IT attacks include the late November ransomware attack on Ardent Health Services, which diverted ambulances and affected health emergency services across multiple U.S. states, and the attack on a water system in western Pennsylvania — claimed by an anti-Israeli Iranian cybercriminal group.

Organizations operating OT-IT systems must modernize legacy technology, deploy layered security, segment IT and OT networks, and implement robust access controls to prevent attacks.

3. Dark Web

The Dark Web, a hidden portion of the internet accessible only through specialized software and configurations, is a breeding ground for illegal activities. New trends on the dark web include the rise of organized criminal activity, characterized by the availability of:

• No-code malware, which requires minimal technical expertise to use.
• Plug-and-play kits, which are pre-configured tools for launching cyberattacks.
• Dedicated customer support.

Additionally, fileless attacks, where attackers use stolen credentials purchased on the Dark Web to gain access to systems without leaving behind traditional malware traces, are one of the biggest trends to look out for. And zero-day brokers — cybercrime groups selling zero-day exploits on the Dark Web to multiple buyers — are becoming increasingly prevalent.

SEE: Here’s everything you need to know about the Dark Web.

In light of these evolving threats, it is crucial for organizations to consider actively monitoring the Dark Web through professional services. This proactive approach can provide valuable insights to help organizations mitigate the great number of different threats that come directly from the Dark Web.

4. Malware as a service and hackers-for-hire

The MaaS landscape has seen a dramatic increase in the availability of platforms and tools that broaden the range of accessible malware and attack functionalities. MaaS user interfaces have also become increasingly intuitive, incorporating tutorials and simplified processes, and diversified. They now cater to various budgets and needs that further lower the barrier to entry, while automation features have become increasingly prevalent.

Meanwhile, hackers-for-hire has become the norm, going even beyond the trend of effectively lowering the technical barriers of launching cyberattacks. This democratization of cybercrime is predicted to fuel a surge in both the number and sophistication of attacks in 2024. According to a Kaspersky report, 2024 will see more groups offering hack-for-hire services. Meanwhile,

SEE: A Kaspersky report reveals the top cyber threats for SMBs in 2023.

To navigate this evolving threat landscape, organizations must prioritize implementing strong layered security solutions capable of detecting and blocking malicious software before it can take root. By equipping employees with knowledge about MaaS and hackers-for-hire threats and social engineering tactics used to distribute malware, organizations can build a more resilient workforce. Regular data backups and encryption, coupled with a zero-trust security model, further bolster defenses by minimizing potential data loss and ensuring stringent access controls.

5. Modern phishing

Phishing attacks that leverage social engineering techniques and personalized messages to trick victims into revealing sensitive information or downloading or clicking on malicious files is evolving.

Traditional methods like mass-mailed generic messages are giving way to personalized and highly realistic attacks. Criminals use AI to automate campaigns and personalize messages with targeted details, generate convincing content like deep fakes and even automatically learn from successes.

To stay ahead, organizations must invest in tools that can detect AI-generated content, educate employees about these evolving threats, and run phishing simulations to identify the weak points in their organizations and secure workplaces.

6. IoT and Industrial IoT

IoT and Industrial IoT devices, with their growing ubiquity and often limited security, present an increasingly attractive target for cybercriminals. In 2023, attacks on IIoT devices saw a significant rise, with attackers leveraging vulnerabilities to launch distributed denial-of-service attacks, steal data and disrupt operations. These attacks evolved to include new techniques like exploiting supply chain vulnerabilities and compromising firmware updates, highlighting the need for enhanced security measures.

SEE: Top IIoT security risks.

To protect against these evolving threats in 2024, organizations must prioritize robust security practices throughout the entire IoT ecosystem. This includes implementing secure coding practices, regularly updating software and firmware, utilizing strong authentication protocols, and monitoring networks for suspicious activity.

Additionally, organizations need to consider adopting zero-trust security models and implementing segmentation strategies to isolate compromised devices and minimize attack impact.

7. State-sponsored attacks

Nation-state actors are increasingly using cyberattacks to achieve their political and strategic goals. These attacks can target critical infrastructure, steal sensitive information and disrupt essential services. 2023 saw an escalation of nation-state-supported cyber criminal activity linked to North Korea, seeking new mechanisms to fund weapon and government programs and navigate international sanctions; Russia, with hackers supporting the invasion of Ukraine and taking cyber warfare to international levels; and the new Israel-Hamas conflict.

Building strong relationships with government and law enforcement agencies and reporting security incidents is fundamental for organizations to mitigate state-backed threats.

2024 demands a proactive approach to thwarting state-sponsored attacks. Organizations need multilayered defenses, including sophisticated cybersecurity solutions, threat intelligence monitoring and robust incident response plans. By prioritizing comprehensive defense strategies and collaborating across sectors, organizations can better protect themselves from the evolving tactics of nation-state actors.

DOWNLOAD: These may be the top threats for 2024, but here are 50 cybersecurity threats to watch out for.

Staying vigilant in the evolving threat landscape

The cybersecurity landscape is constantly evolving, and threats are becoming more sophisticated. To mitigate modern cybersecurity and compliance threats, organizations must combine state-of-the-art technologies operating under holistic cybersecurity programs.

Strategies like zero-trust models are essential to strengthening companies’ security postures as they adapt efficiently and proactively to cybersecurity threats. Kolide — which sponsored this forward-looking report — offers a user- and device-level trust solution that empowers organizations with Okta to seamlessly deploy zero-trust access models and secure their environment and apps.

By staying vigilant and adapting to the changing threat landscape, organizations can protect themselves from cyberattacks and ensure the security of their data and systems.

Learn about Kolide

Although AI has been around since the 1950s, it has seen tremendous growth within the past year. Tech giants have been implementing AI into their products and services, while individuals are using it to make their lives a little easier.

Deloitte surveyed companies and professionals in its second edition of the "State of Ethics and Trust in Technology" report, led by its Technology Trust Ethics practice. According to the report, 74% of companies have already begun testing generative AI, while 65% have begun to use it internally. The increasing awareness of AI's new capabilities has led to the pressing question of how organizations can use this technology ethically.

Also: The ethics of generative AI: How we can harness this powerful technology

Deloitte interviewed 26 specialists in various industries to gather information about how industry leaders are considering concerns about the ethical use of emerging technologies, including generative AI.

The company then tested hypotheses and delivered a 64-question survey to more than 1,700 businesses and technical professionals to gain further insights.

The report, by Beena Ammanath, managing director of Deloitte Consulting LLP and leader of Deloitte's Technology Trust Ethics practice, refers to emerging technologies as the following: Cognitive technologies (including general and generative AI and chatbots), digital reality, ambient experiences, autonomous vehicles, quantum computing, distributed ledger technology, and robotics.

According to the survey, 39% of survey respondents, consisting of business leaders and developers of emerging technologies, thought cognitive technologies had the most potential for social good, compared to 12% in digital reality, and 12% in ambient experiences.

Also: 5 essential traits that tomorrow's AI leader must have

However, 57% of survey respondents also thought that cognitive technologies had the greatest potential for serious ethical risk.

The most concerning statistic is that over half of the respondents (56%) said their "company does not have or are unsure if they have ethical principles guiding the use of generative AI."

Compared to Deloitte's report in 2022 about ethics and trust in emerging technologies, this year's report reveals that "organizations find themselves wrestling with new ethical issues posed by wide-scale adoption of this once-again new technology."

These issues are tied to concerns about how businesses and organizations are using these technologies.

Despite the many benefits of AI, 22% of respondents were concerned with data privacy while 14% cited transparency about how AI is trained with data to produce its outputs.

Also: Does your business need a chief AI officer?

Data poisoning as well as intellectual property and copyright were concerns that each consisted of 12% of survey respondents. Data poisoning is the "pollution" of data training sets by bad actors and can lead to inaccurate results produced by AI.

Deloitte's report also detailed the types of damage that survey respondents believe could arise when ethical violations are not taken seriously.

Reputational damage was the greatest source of concern coming from 38% of respondents, followed by human damage such as misdiagnoses or data privacy violations (27%), regulatory penalties like copyright infringement (17%), financial damage (9%), and employee dissatisfaction (9%).

These damages are evident in the several lawsuits that have already been filed due to privacy violations, copyright infringement, and other issues related to the unethical use of AI.

Also: AI and automation: Business leaders adopt small-scale solutions for greater impact

So how can companies ensure they using AI safely? Deloitte lists a multi-step approach to helping companies:

• Exploration: Companies can begin by letting product owners, business leaders, and AI/ML practitioners explore generative AI through workshops to see how it could create value for their businesses. This way, companies can recognize the costs and benefits of incorporating AI into their businesses.
• Foundational: Companies could buy or build AI platforms to implement generative AI into their businesses. Of the survey respondents, 30% of survey respondents' companies chose to use existing capabilities with major AI platforms. 8% of respondents created their own in-house AI platforms, while 5% decided not to use generative AI.
• Governance: Creating standards and protocols for AI use could minimize the potentially harmful impacts of AI, so companies should determine what types of ethical principles they plan to uphold.
• Trainings and education: Companies could mandate trainings that outline the ethical principles of using AI. In addition, technical trainings that educate employees about using a variety of LLMs could provide companies with more guidance about the ethical use of AI.
• Pilots: Engineers and product leaders could run experiments on a variety of use cases to test proof of concepts and pilot programs and then eliminate aspects that are too risky.
• Implementation: Companies should draft a plan for introducing a newly enhanced product into the market and assign accountability for product implementation and ownership. The company should also have a team of experts prepared to address any issues that may arise. Transparency is also crucial for this step, as companies should explain how user data is inputted into the model, how the model reaches its output, and how likely the model is to hallucinate.
• Audit: According to one interviewee, companies will need to modify their policies depending on the risks of AI use. This could vary company by company, as not all organizations will incorporate AI for the same use case.

In considering the impact of generative AI on human workers, issues such as transparency and data privacy ranked above job displacement. Nevertheless, the report also mentioned that "49% said workers at their organization displaced by AI moved to different roles and retrained and upskilled." Furthermore, 11% were terminated, 13% were put in a different role without being retrained or upskilled, and 27% did not experience any job displacement at their organization from AI, according to Deloitte.

Also: Will AI hurt or help workers? It's complicated

"The sooner companies work together to identify the risks and establish governance up front, the better their ability may be to help generate stakeholder value, elevate their brands, create new markets, and contribute to building a more equitable world," said Ammanath.

Artificial Intelligence

Believe it or not, there is still a shortage of NVIDIA GPUs in the market. AI startups, along with Fortune 500 companies, have been struggling to onboard GPUs for training AI models, even as small as Llama 7B. This is where Lamini with its partnership with AMD has found its moat — LLM fine tuning on AMD using ROCm.

“If you are an AI startup blocked on GPUs, send me a note,” posted Gregory Diamos, co-founder of Lamini, on LinkedIn. “At Lamini, we have figured out how to use AMD GPUs, which gives us a relatively large supply compared to the rest of the market.”

Similarly, co-founder Sharon Zhou posted on X, “We just brought online a relatively enormous supply of GPUs Lamini. We’re allocating a significant subset to promising startups & open-source initiatives, building the future of LLMs.”

Lamini co-founders claimed that they have the most competitive perf/$ on the market right now, “because we figured out how to use AMD GPUs to get software parity with CUDA, trending beyond CUDA.”

Zhou, at AMD’s Advancing AI event, also discussed how they have been leveraging AMD hardware and software all this while, and proving that the open nature of the technology has been helping them fully own the technology. “We have reached beyond CUDA,” said Zhou. This was after the big reveal two months ago that Lamini was exclusively running on AMD GPUs for the past year.

What exactly has Lamini figured out?

In September, Lamini opened up its LLM Superstation for both, cloud and on-premise for its customers. The LLM Superstation is a finely tuned supercomputer that incorporates 128 AMD Instinct GPUs, utilising Lamini on the AMD ROCm open software ecosystem.

Read: AMD’s ROCm is Ready To Challenge NVIDIA’s CUDA

Lamini and ROCm have achieved significant maturity, facilitating the effective fine-tuning of expansive LLMs, including Meta AI’s Llama 2, which now Lamini customers can book for training LLMs, and making their AI models proprietary.

To ensure this, Lamini incorporates sophisticated optimisations tailored for enterprise LLMs, leveraging and extending PEFT (LoRA), RLHF, and toolformer. These optimisations enable data isolation across 4,266x models on a single server, accelerate model switching by 1.09 billion times, compress models by a factor of 32x, and seamlessly integrate LLMs with enterprise APIs without the need for hyperparameter search.

Furthermore, Lamini has integrated a range of novel optimisations designed to expedite LLMs while harnessing the distinctive capabilities of AMD’s MI platform. These enhancements empower the hosting of 200 billion parameter models on a singular server, accommodating 10,000 fine-tuned language models on a single server, managing 12,800 concurrent requests to a solitary server, and efficiently processing more than 3.5 million queries per day on a single node.

“What’s more, with Lamini, you can stop worrying about the 52-week lead time for NVIDIA H100s,” reads the Lamini blog. “Using Lamini exclusively, you can build your own enterprise LLMs and ship them into production on AMD Instinct GPUs.” The company claims the cost of using Lamini on AMD Instinct GPUs is 10 times lesser than AWS, without the wait time.

“We’ve deployed Lamini in our internal Kubernetes cluster with AMD Instinct GPUs, and are using fine tuning to create models that are trained on AMD code base across multiple components for specific developer tasks,” said Vamsi Boppana, SVP of AI at AMD.

Moreover, Diamos claims that MI250X runs bigger models than NVIDIA’s A100s. “We chose the Instinct MI250 as the foundation for Lamini because it runs the biggest models that our customers demand and integrates fine tuning optimisations. We use the large HBM capacity (128GB) on MI250 to run bigger models with lower software complexity than clusters of A100s.”

Companies love AMD GPUs

“Building LLMs should be easy. Every enterprise should be able to own LLM IP, just like they do for all their other software. We’re excited to partner with AMD because their GPUs unlock a huge opportunity for enterprises to get started with little to no lead time,” said Zhou. “The main reason for this is compute strategy.”

Read: AMD Loves Llama So Much

Apart from Lamini, and through Lamini, many companies are increasingly testing out AMD’s products. Now that MI300X is also out, companies such as Microsoft, Meta, and Oracle have already announced the integration of the GPUs into their data centres and providing their customers AMD compute. OpenAI is also integrating support for ROCm on its Triton compiler.

Databricks and Essential AI had also announced the use of MI250X for their services and said at the Advancing AI event that they are ready for MI300X.

All this is because of AMD’s performance and its open source approach. The training performance of the MI300X is exactly equal to the NVIDIA H100. But when it comes to inference, MI300X using Bloom 176B and Llama 2 70B offers 1.6X and 1.4X faster performance.

Zoho’s Sridhar Vembu also revealed at the Global AI Conclave that given the shortage of GPU supplies, the company has also been running its models on AMD. “AMD has a very competitive silicon now, so with that I think the supply situation should resolve then it’s a matter of getting the talent capital.”

The post AMD and Lamini Come to the Rescue of AI Startups appeared first on Analytics India Magazine.

Llama 2 is currently one of the sought-after open-source models globally, with 757K downloads on HuggingFace. Its free availability for research and commercial purposes make it attractive to a broad user base, including researchers, developers, and hobbyists.

It has gained popularity in India and other countries worldwide, where users have created ‘Local Llamas’ based on the local language of the region. In models like GPT-3.5, there is a challenge with ‘tokenisation,’ where Indic language text is not efficiently represented.

Compared to massive LLMs like GPT-4 or 3.5, Llama 2 requires less computational power and training data. This makes it more feasible to train and run on local hardware, even in areas with limited infrastructure.

Moreover, Llama 2 can be easily fine-tuned for specific tasks and domains using smaller datasets of local language text. This allows developers to tailor the model to their specific language and needs, improving its performance and relevance.

OpenHathi

Recently in India, Sarvam AI introduced OpenHathi-Hi-v0.1, marking the debut of the first Hindi LLM in the OpenHathi series. Built on a cost-effective platform, this model, extending from Llama2-7B, exhibits performance akin to GPT-3.5 for Indic languages.

OpenHathi, featuring a 48K-token extension of Llama2-7B’s tokenizer, undergoes a two-phase training process. The initial phase focuses on embedding alignment, aligning randomly initialized Hindi embeddings, followed by bilingual language modeling, teaching the model cross-lingual attention across tokens.

The model demonstrates robust performance across various Hindi tasks, comparable to, if not surpassing, GPT-3.5, while maintaining English proficiency. Sarvam AI’s evaluation includes non-academic, real-world tasks alongside standard Natural Language Generation (NLG) tasks. Evaluations against GPT-3.5 generation with GPT-4 as the judge revealed superior performance in Hindi, both in native and Romanised scripts.

Tamil Llama

Kaggle ML Engineer and Kaggle Master Abhinand Balachandra, recently introduced Tamil Llama, an Indic LLM engineered specifically to elevate the Tamil language domain. This AI model is built on top of Meta’s Llama 2.

It was trained with an additional 16,000 Tamil tokens, aiming to achieve superior text generation and comprehension in the Tamil language. This model serves as an extension of the LLaMA model and has been enhanced by incorporating extra Tamil tokens, utilizing the LoRA methodology for efficient training.

It presents four distinct variations: Tamil LLaMA 7B, 13B, 7B Instruct, and 14B Instruct. Throughout the training phase, the model’s vocabulary has been expanded to encompass 16,000 Tamil tokens, supplementing the original 32K tokens.

Telugu Llama

Ramsri Goutham Golla from Segmind.com is currently working on Telugu Llama. Meticulously trained on LLama 2, the popular language model, showcasing remarkable efficiency in token count for Telugu text.

Interestingly, Telugu language consumes less tokens as compared to English as per Golla. This breakthrough not only promises faster and cost-effective text generation in Telugu but also sets the stage for LLama 2 to excel in Indic languages, revolutionising the landscape of natural language processing.

Odia Generative AI

odia_llama2_7B_v1 created by Odia Generative AI is based on Llama2-7b and fine tuned with a 180k Odia instruction set. This set includes translated data from open-source resources and a purposefully crafted domain knowledge instruction set. The result is a model that effectively understands Odia instructions and generates responses, demonstrating its practical utility for the nuances of the Odia language.

SeaLLMs – Large Language Models for Southeast Asia

Alibaba Group Holding’s research division, Damo Academy, has introduced LLMs specifically designed for Southeast Asian languages. SeaLLMs are built upon the Llama-2 model and further advanced through continued pre-training with an extended vocabulary, specialized instruction and alignment tuning to better capture the intricacies of regional languages.

The Southeast Asia LLM (SeaLLM) was pre-trained on Vietnamese, Indonesian, Thai, Malay, Khmer, Lao, Tagalog, and Burmese data sets, and has outperformed other open-source models in linguistic and safety tasks. SeaLLMs exhibit remarkable proficiency in language understanding and generation tasks, presenting a formidable challenge to dominant models like ChatGPT-3.5, especially in Southeast Asian (SEA) languages.

VinaLLaMA: LLaMA-based Vietnamese Foundation Model

VinaLLaMA, a foundational LLM designed specifically for the Vietnamese language. VinaL- LaMA, built on top of LLaMA-2, represents a vital stride towards linguistic inclusivity in AI, adeptly addressing the syntactic and semantic intricacies of Vietnamese.

LLaMAntino: LLaMA 2 Models for Effective Text Generation in Italian Language

LLaMAntino is the Italian iteration of Llama 2, developed by researchers at the University of Bari Aldo Moro, Italy. Through the fine-tuning of pre-trained LLaMA 2 models with 13 and 7 billion parameters and using a substantial dataset of Italian text, they created the LLaMAntino-2-7b-hf-ITA and LLaMAntino-2-13b-hf-ITA models.

This model aims to provide Italian NLP researchers with a tool to tackle tasks such as information extraction and closed qa.

Dutch Llama v2 13B

It is a Dutch model based on Llama 2. This model is a fine-tuned version of BramVanroy/llama2-13b-ft-mc4_nl_cleaned_tiny on the BramVanroy/dutch_chat_datasets dataset on a context of 4096 tokens.

Depending on the input, the model can provide satisfactory outcomes, taking into account its 13B size and limited pretraining in Dutch. However, it’s essential to note that the model lacks human feedback during training and lacks safeguards, potentially leading to unexpected or offensive content based on the query.

The post Top 8 Vernacular Language Models Based on Llama 2 appeared first on Analytics India Magazine.

Tech Mahindra’s outgoing chief, CP Gurnani, shared a Project Indus update on X, highlighting the development of an LLM specifically designed for Hindi and its 37 dialects.

Retiring after 19 years at Tech Mahindra he said, “I feel proud to share that the GenerativeAI project challenge I took up earlier has been successfully accomplished by our brilliant research team at Makers Lab.” Project Indus is in the beta testing phase within TechM. The pure Hindi LLM consists of 539 million parameters and 10 billion Hindi+ dialect tokens, as shared by Gurnani.

“The model is probably the only one in the world that has all Hindi tokens and has been trained from the ground up. It will set the stage for the years to come as our prowess in deep tech,” he said.

“I now pass the baton over to Mohit Joshi, Nikhil Malhotra, and the team, as well as my talented Tech Mighties, who will take this one notch higher. Thank you for everything, team!” he added.

In October, Tech Mahindra announced its plan to release Project Indus by the end of December or early January. Introduced in August, the model is initially set to support 40 different Hindi dialects, with plans to add more languages and dialects in subsequent releases. Over the last two months, the 15-member Project Indus team has collected 1.2 terabytes of data in Hindi and related dialects.

The update by Gurnani comes in the backdrop of last week, which witnessed a slew of announcements from Indian companies and startups launching their large language models (LLMs). This includes Google-backed CoRover’s BharatGPT, Khosla Ventures-backed Sarvam.ai’sOpenHathi, Microsoft-backed Kissan AI’s Dhenu, and Ola’s Krutrim.

The post Tech Mahindra Completes Project Indus, to Launch Soon appeared first on Analytics India Magazine.