Apple acquires AI startup specializing in overlooking manufacturing components Ivan Mehta 8 hours
Apple has added another AI startup to its acquisition list with Canada-based DarwinAI, which specializes in vision-based tech to observe components during manufacturing to improve efficiency, Bloomberg reported.
While Apple or DarwinAI haven’t announced this deal, several members of the startup’s team joined Apple’s machine learning teams in January, as per their LinkedIn profiles.
DarwinAI had raised over $15 million in funding across various rounds from investors including BDC Capital’s Deep Tech Venture Fund, Honeywell Ventures, Obvious Ventures, and Inovia Capital. BDC Capital confirms on its website that it has received an exit from DarwinAI, whereas Obvious Venture has updated its portfolio to reflect that the startup has been acquired.
BDC Capital and Obvious Venture didn’t comment on the story at the time of writing. Apple didn’t immediately respond to requests for comment.
As Bloomberg noted in its report, apart from helping with manufacturing efficiency, DarwinAI uses techniques to make AI models smaller and faster. This could be useful for on-device generative AI features Apple hopes to introduce in iOS 18 this year.
Apple has lagged in releasing gen-AI-powered features as compared to competitors like OpenAI, Google, Meta, and Microsoft. In February, Tim Cook said the company plans to introduce such features “later this year.”
“We continue to spend a tremendous amount of time and effort and we’re excited to share the details of our ongoing work in that space later this year,” Cook said during a quarterly earnings call with analysts.
Last year the Apple top exec had confirmed investment into gen-AI efforts. Apple’s job listings have suggested the company is exploring infusing AI in multiple internal and external areas including, Siri, developer tools, and customer support.
Google has already had a pretty eventful year, rebranding its AI chatbot from Bard to Gemini and releasing several new AI models. At Google I/O, we expect the company to make even more announcements regarding AI, and unveil new hardware for the first time this year.
Google sent out event invites to the press — including ZDNET — which revealed that the annual developer conference is taking place on May 14 and 15 at the Shoreline Amphitheatre in Mountain View, CA.
Also: Google Messages rolls out iMessage-like emoji reactions: Here's how to use them
The keynote event, which is when Google leaders take the stage to unveil the company's latest hardware and software, is going to take place on the first day of the conference at 10 AM PT / 1 PM ET.
Although the invite doesn't specify details about video streaming, you can expect the company to live stream the event on YouTube and its website as it usually does for members of the public and press who can't attend in person.
Developers can register for the event on the Google I/O landing page for free to take advantage of perks such as getting email updates and accessing on-demand sessions.
Google's event description on both the invite and the website was vague. The invite read, "Learn about Google's newest developer tools and discover how they fuel innovation and enhance your workflow for maximum productivity."
However, given the announcements already made this year, we expect the focal point of the event to be artificial intelligence. At last year's I/O, AI was also the crown jewel of the event with major announcements including two new large language models (PaLM 2 and Gemini), new AI features in Android 14, an improved chatbot, and more.
Similarly this year, you can expect to learn more about the new features on Android 15, which will likely feature a myriad of new AI capabilities, updates to Gemini (formerly Bard), and maybe even a new large language model.
Also: The best Android phones to buy in 2024
The event will presumably showcase some new hardware too. If Google follows last year's precedent, it will unveil a new smartphone, the Pixel 8A. The Pixel 7A, revealed at least year's event, proved to be an impressive A-series device and mid-range smartphone when reviewed by ZDNET.
Google I/O is notorious for throwing surprises, so it is a good rule of thumb to expect something no one is talking about. ZDNET will be covering the event, so even if you can't tune in, you can check the site for the latest updates.
Zscaler buys Avalor to bring more AI into its security tools Kyle Wiggers 10 hours
Zscaler, a cloud security company with headquarters in San Jose, California, has acquired cybersecurity startup Avalor 26 months after its founding, reportedly for $310 million in cash and equity.
In a press release announcing the news, Zscaler founder and CEO Jay Chaudhry said that the deal would expand Zscaler’s platform with capabilities including streamlined reporting of security incidents, incident mitigation, asset discovery, data classification, security policy generation and more.
“AI is only as good as the underlying data, and many solutions lack the additional context and knowledge from data sources across the enterprise to truly leverage security-specific AI models,” Chaudhry said in a press release. “Zscaler operates the world’s largest security cloud with the most relevant data to train security specific large language models, and with the Avalor acquisition, we can more effectively identify vulnerabilities while predicting and preventing breaches.”
Raanan Raz co-founded Avalor with Kfir Tishbi, who previously led the engineering team at Datorama, a marketing analytics company acquired by Salesforce in 2018. Raz and Tishbi worked together at Datorama both leading up to — and after — the Salesforce purchase.
Avalor acts as a source of truth for cybersecurity assets, controls, identities, vulnerabilities, bugs and other data points, allowing security teams to aggregate, normalize, de-duplicate and track risk data from discovery to remediation.
It’s not a unique concept. An array of startups out there tackle the same problem, like Securiti and Dig Security. But what sets Avalor apart is the ability to handle data from virtually any source in any format, and its unique set of vulnerability risk management and prioritization tools.
Prior to the Zscaler acquisition, Avalor managed to secure $30 million from investors, including TCV, Salesforce Ventures, Jibe Ventures and Cyberstarts. And Raz sees Zscaler taking the business — and its ~80-person team spread across the U.S. and Israel — further.
“[With Zscaler, ] we get instant access to a set of resources it would have taken years for us to develop organically — 7,000 customers, 4,200 channel partners globally, near-ubiquitous customer awareness and the validation of a $2 billion business behind us,” Raz wrote in a post on Avalor’s blog published Thursday morning. “We’ll continue to operate independently as a complete Avalor team, and we’ll have all the tailwinds of the amazing Zscaler resources helping us.”
Avalor is Zscaler’s third acquisition after Canonic, a startup focused on protecting against cyberattacks targeting software-as-a-service products, and Trustdome, a cloud infrastructure entitlement platform. Founded in 2007 by Chaudhry and K. Kailash, Zscaler — which went public in March 2018 — has ~7,000 employees and a market cap of about $30 billion.
As Crunchbase’s Chris Metinko noted earlier today, Zscaler’s acquisition — along with others in the cybersecurity space — could help spark activity in a slow-to-stagnant cyber M&A market. Last year saw only 66 M&A deals involving VC-backed cybersecurity startups, per Crunchbase — a 26% drop from 2022 (which saw 89 such deals) and more than a 50% decline from 2021 (139 deals).
So far in 2024, there have been 18 cybersecurity-related mergers and acquisitions, with especially notable moves being made by Wiz, SentinelOne and CrowdStrike.
In a recent interview with the Wall Street Journal, CTO Mira Murati shared that OpenAI will make its latest text-to-video generator, Sora, publicly accessible later this year. When Sora was launched earlier in February, users greatly appreciated its hyper-realistic videos, many calling it the “ChatGPT moment for video.”
The model, showcased in February, generates realistic scenes from text prompts and will soon be open for public use. The initial rollout will primarily target visual artists and filmmakers. Murati also disclosed plans to incorporate sound and editing flexibility into Sora-generated videos.
Sora was trained on publicly available and licensed data, including content from Shutterstock. In July, the company expanded its partnership with OpenAI, signing a new six-year agreement to provide high-quality training data.
However, she mentioned that given that Sora is “much more expensive”, the team is trying to make it financially viable, similar to its other products like DALL.E and ChatGPT.
The team behind Sora is led by Tim Brooks, a research scientist at OpenAI; William Peebles, a research scientist at OpenAI, and Aditya Ramesh, the creator of DALL·E and the head of Videogen.
OpenAI’s all-new super-cool, text-to-video tool can create videos of up to 60 seconds featuring highly detailed scenes, complex camera motion, and multiple characters with vibrant emotions. Unlike typical language models using text tokens, Sora utilizes visual patches, similar to LLMs, focusing on visual elements.
Sora is a stepping stone for AGI, as the company teaches the model to understand and simulate real-world interactions. Today marks one year of GPT-4. It is time for OpenAI to come up with GPT-5.
The post OpenAI to Make Sora Publicly Accessible Later this Year appeared first on Analytics India Magazine.
Most attempts at building a humanoid robot, such as Tesla's Optimus, focus on assisting humans with physical, manual tasks. Figure is among the AI robotics startups that aren't satisfied with just movement. Figure is trying to take its humanoid robots to the next level by integrating language—and the results are quite impressive.
On Wednesday, Figure posted a two-minute video on X to demo what its humanoid robot Figure 1 can purportedly do thanks to a combination of the startup's neural networks and OpenAI's visual and language intelligence. Watch the video below.
Also: How work devices may evolve to keep up in the AI era
The video opens with a person asking Figure 1 what the robot sees, to which Figure 1 conversationally explains exactly what is on the table in front of it, showcasing the robot's advanced natural language processing, language, and visual capabilities.
Then, perhaps even more impressively, when the person asks Figure 1 for something to eat, the robot responds by choosing the apple amongst the items in front of it, highlighting the robot's ability to reason through a situation and deliver based on that processing.
When prompted, Figure 1 was even able to explain its choice, the video shows, while performing another task saying, "So I gave you the apple because it is the only edible item I could provide you with from the table."
Also: Microsoft quietly upgraded Copilot's free version to GPT-4 Turbo. Here's why it matters
Figure announced its OpenAI partnership at the end of February with the hopes of leveraging OpenAI's research to enhance its humanoid robots' capabilities to process and reason from language. The robotics startup hopes the collaboration will accelerate how quickly its robots can come to market.
While it's easy to curate a demo like this to be as impressive as possible, not to mention edit any resulting footage, Figure appears to be making strides towards its goal. If this video is an accurate representation of what Figure 1 can do, Figure could be on the brink of a major breakthrough in the robotics space—a commercially available robot that can not only perform physical tasks, but also reason and speak, too.
Microsoft has announced the general availability of Microsoft Copilot for Security, set to debut on April 1, 2024. This groundbreaking AI-driven solution, a first of its kind in the industry, is poised to revolutionise the capabilities of security and IT professionals by providing them with enhanced tools to detect threats, expedite response times, and augment team expertise.
Excited to announce this milestone for Copilot for Security, as we continue to apply this next generation of AI to empower defenders across all experience levels and transform every aspect of SOC productivity. https://t.co/2IMKcAs0OC
— Satya Nadella (@satyanadella) March 13, 2024
Powered by vast volumes of data and threat intelligence, including over 78 trillion security signals processed daily by Microsoft, Copilot leverages advanced language models to offer tailored insights and actionable guidance. This innovative approach empowers users to safeguard their systems at the rapid pace and scale facilitated by artificial intelligence, thus transforming the landscape of security operations.
In a bid to make Copilot for Security more accessible to a diverse array of organisations, Microsoft has introduced a provisioned pay-as-you-go licensing model. This flexible pricing structure enables swift deployment and allows organisations to scale their usage and costs in alignment with their specific needs and budgetary constraints.
Furthermore, the global rollout of Copilot for Security underscores its multilingual capabilities, with support for eight languages in prompt processing and response, alongside a multilingual interface catering to 25 different languages. This broad linguistic coverage ensures readiness for deployment across major regions spanning North and South America, Europe, and Asia.
Comprising over 100 partners, including managed security service providers and independent software vendors, these partnerships play a pivotal role in facilitating widespread adoption of AI-driven security solutions and empowering organisations to navigate the evolving threat landscape with confidence.
As part of its general availability, Copilot for Security introduces several new capabilities, including custom promptbook creation, knowledge base integrations, and third-party integrations, aimed at enhancing operational efficiency and threat mitigation strategies.
Microsoft is equipping customers with enhanced visibility, protection, and governance mechanisms. These initiatives encompass the discovery of AI risks, safeguarding of AI applications and data, and governance of AI usage, ensuring secure and responsible adoption of AI technologies.
Moreover, Microsoft continues to bolster its comprehensive suite of security offerings, with recent additions such as Microsoft Security Exposure Management and Adaptive Protection integrated with Microsoft Entra Conditional Access. These enhancements underscore Microsoft’s unwavering commitment to providing customers with end-to-end protection for their digital assets, bolstering resilience against emerging threats.
The launch of Microsoft Copilot for Security heralds a new era in cybersecurity, empowering organisations worldwide to fortify their defences, respond swiftly to evolving threats, and embrace the transformative potential of artificial intelligence in safeguarding digital assets.
The post Microsoft Copilot for Security to be Generally Available Soon appeared first on Analytics India Magazine.
Google Sheets users can now use Claude 3, Anthropic’s latest LLM, surpassing GPT-4 on prominent benchmarks, including near-instant results and strong reasoning capabilities. With Claude 3, users can streamline their workflow by creating prompt templates within their spreadsheets and filling them with customised data.
How to use Claude 3 in Google Sheets
X user Moritz Krembshared how to use Claude 3 on sheets. First, users need to install the Claude For Sheets extension, available on the Google Workspace Marketplace. Once installed, they get an API key from the Anthropic Console. With the API key in hand, you have to call Claude using the formula: =Claude(prompt, model)
Once configured, Claude 3 offers a range of possibilities for enhancing productivity and efficiency. A primary feature is its ability to generate personalised content based on user-defined prompts and input data. For example, if you want to craft personalised cold emails tailored to specific customers, you can integrate customer data such as names, industries, and products.
With the ability to automate repetitive tasks and generate tailored content swiftly, Claude 3 empowers users to focus their time and energy on high-value activities, improving productivity.
The Claude 3 model series debuts with a 200,000-token context window, a jump from 100,000 tokens in the second version of Claude. However, these models are flexible in accommodating inputs surpassing one million tokens for selected customers.
Opus is the flagship model of the Claude 3 family. Meanwhile, Claude 3 Sonnet ( available for free now) is designed for enterprise workloads, and Claude 3 Haiku stands out as the fastest and most compact model, ensuring near-instant responsiveness.
Yesterday, Amazon announced that Claude 3 Haiku is now available on Amazon Bedrock.
The post This is How You Can Use Claude 3 in Google Sheets appeared first on Analytics India Magazine.
Amazon shoppers are used to sifting through thousands of product listings to find exactly what they want, and most sellers put in a lot of work to ensure that each listing is as detailed and accurate as possible. With the help of artificial intelligence (AI), this will soon become easier than ever.
Amazon has announced a feature that lets sellers convert their existing product pages from other websites into detailed Amazon listings by simply copying and pasting a link. Designed to streamline the listing creation process, the feature is powered by AI through Amazon's cloud computing service, AWS, and leverages proprietary foundational models available on Amazon Bedrock.
Also: The 45+ best Amazon Big Spring Sale deals you can shop now
Sellers will be able to enter a link to a product listing from their own direct-to-consumer website on Amazon, and Amazon's generative AI tools will parse the information and create engaging listings. Amazon says the new capability will roll out gradually and be available to US sellers within the coming weeks.
"We are finding that generative AI has a profound capacity to simplify the effort it takes for our selling partners to create great product listings, and we will continue to innovate upon these capabilities," explained Mary Beth Westmoreland, Amazon's vice president of worldwide selling partner experience, in a blog post. "As we expand the breadth of attributes that we can automatically generate and further improve the quality of generated content, we keep reducing the effort needed for sellers to create high-quality product detail pages that resonate with customers."
Also: The 40+ best Amazon tech deals of March 2024
The e-commerce giant has been integrating generative AI features into its platform since last fall. These AI tools let sellers generate product titles, descriptions, and details by providing a few descriptive words or images of the product. On the buyer's side, Amazon has integrated AI-generated review summaries that give users a synopsis of what buyers have said in their product reviews.
Accompanying the announcement of this latest feature for sellers, Amazon reported that more than 100,000 sellers have used the generative AI features made available for them.
Google researchers recently introduced Synth2, a novel approach for training Visual-Language Models (VLMs) using synthetic image-text pairs. This new technique enhances the capabilities of LLMs and text-to-image generation, alongside addressing the limitations of manual image labelling.
This method solves the problem of not having enough data labelled by people. It uses LLMs to generate captions and a text-to-image model to synthesise corresponding images.
Google researchers demonstrate several significant findings in terms of improved VLM performance, data efficiency, and customisation and scalability.
For instance, the visual language model trained on their synthetic and human-annotated datasets showed marked improvement in image captioning tasks compared to a baseline trained exclusively on human-annotated data.
“This underscores the potential of our method to augment VLM capabilities efficiently. This is highly advantageous in scenarios where data acquisition and annotation are resource-intensive,” said the researchers.
In addition, this new approach yielded comparable performance while utilising only a fraction of human-labelled data, thereby increasing data efficiency.
This method proves to be flexible in customising generating image datasets for specific domains. The synthetic data generation is also scalable to support large scale VLM development.
You can find the research paper here.
The post Google Introduces Synth^2 to Enhance the Training of Visual Language Models appeared first on Analytics India Magazine.
Cosine similarity, a popular metric in assessing semantic similarity between high-dimensional objects, is facing scrutiny regarding its universal applicability. While commonly employed across various domains, recent studies suggest that its efficacy might not be consistent across all scenarios, prompting a re-evaluation of its necessity.
In a recent paper by Netflix researchers Harald Steck, Chaitanya Ekanadham, and Nathan Kallus, titled ‘Is Cosine-Similarity of Embeddings Really Necessary Everywhere?’, the authors explore the nuanced landscape surrounding cosine similarity within the context of linear matrix factorisation models. Their findings challenge the assumption that cosine similarity is indispensable in all AI applications, advocating for a more discerning approach.
Central to the debate is the realisation that cosine similarity may not always capture semantic similarity as reliably as previously thought. The Netflix study uncovers instances where cosine similarity fails to provide meaningful insights, emphasising the importance of considering alternative measures in certain contexts.
"how cosine-similarity can yield arbitrary and therefore meaningless `similarities.'" – New Netflix Paper pic.twitter.com/KVQMONcSIN
— 1LittleCoder (@1littlecoder) March 11, 2024
In natural language processing, for example, where cosine similarity is often used to assess the semantic similarity between words or documents, researchers may benefit from considering alternative metrics that capture the nuances of language semantics better.
Similarly, in recommender systems, where cosine similarity is employed to measure the similarity between users or items, exploring alternative techniques may lead to more accurate recommendations and improved user satisfaction.
What to do?
“Why would anyone expect cosine-similarity to be a useful metric?,” said a user on HackerNews. “In the real world, the arbitrary absolute position of an object in the universe (if it could be measured) isn’t that important, it’s the directions and distances to nearby objects that matter most,” he added.
In traditional linear matrix factorisation models, the reliance on cosine similarity as a default similarity measure is commonplace. However, as the Netflix study reveals, this may not always be justified.
Bindu Reddy, the founder of Abacus.AI, posted on X saying, “It goes back to building good RAG systems, which is hard. Before deploying these systems, you have to make intelligent decisions about chunking, hierarchical chunking, embedding, and even the algorithm for similarity look-up. Failure modes will be high and accuracy low if you don’t use the appropriate techniques.”
Good paper by Netflix on cosine similarity. It goes back to building good RAG systems, which is hard. Before deploying these systems, you have to make intelligent decisions about chunking, hierarchical chunking, embedding, and even the algorithm for similarity look-up.… pic.twitter.com/myUFug7Cag
— Bindu Reddy (@bindureddy) March 11, 2024
A recent blog by Andew Nguonly also suggested adding keyword search to improve the results, and further improving it by involving RAG. “In fact, I’d bet that this is the expectation of many users of RAG applications already,” he said in the blog.
To make this clear, authors of the paper proved it by leveraging analytical derivations from linear matrix factorization models, the authors shed light on the limitations of cosine similarity and propose alternative metrics for similarity evaluation.
Their findings highlight the need for a nuanced understanding of the underlying data and the suitability of different metrics based on the specific requirements of each application, which leads back to using vector embeddings and RAG.
One alternative metric gaining traction in similarity evaluation is the unnormalised dot-product between embedded vectors. Unlike cosine similarity, which relies on the normalisation of vectors, the unnormalised dot-product preserves the original magnitudes of the vectors, providing a different perspective on similarity assessment.
Additionally, techniques based on optimal transport have emerged as promising alternatives to cosine similarity, particularly in scenarios where the underlying data exhibits complex geometric structures. While computationally expensive, these techniques offer a more theoretically sound approach to measuring similarity, taking into account the curvature of the underlying manifold.
Not all that redundant
By embracing a more nuanced approach to similarity evaluation, informed by the latest research and guided by the specific characteristics of their data and analysis goals, researchers and practitioners can unlock new insights and drive innovation in their respective fields. However, the paper also falls short of proposing alternatives to the test, leaving us hanging.
“There aren’t any alternatives: cosine similarity is effectively an extension of Euclidean distance, which is the mathematically correct way for finding the distance between vectors,” explained a user on HackerNews. “You may not want to use cosine similarity as your only metric for rankings, however, and you may want to experiment with how you construct the embeddings.”
While acknowledging the utility of cosine similarity in many scenarios, the authors caution against its indiscriminate use. Instead, they advocate for a case-by-case evaluation of the appropriateness of cosine similarity, considering factors such as data characteristics, model architecture, and the intended use of similarity metrics.
The debate surrounding the necessity of cosine similarity underscores the evolving nature of similarity evaluation in data-driven fields. While cosine similarity remains a valuable tool in many contexts, its blind application may lead to suboptimal outcomes.
Through rigorous analysis and thoughtful consideration of alternatives, researchers and practitioners can navigate the complexities of similarity measurement more effectively, ensuring that the right metric is used for the right task.
The post When (Not) to Use Cosine Similarity appeared first on Analytics India Magazine.