OpenAI is releasing a new flagship generative AI model called GPT-4o, set to roll out “iteratively” across the company’s developer and consumer-facing products over the next few weeks. (The “o” in GPT-4 stands for “omnimodel.”) OpenAI CTO Muri Murati said that GPT-4o provides “GPT-4-level” intelligence but improves on GPT-4’s capabilities across text and vision as […]
© 2024 TechCrunch. All rights reserved. For personal use only.
Airtel, India’s second-largest telco, is partnering with Google Cloud to develop and deliver cloud and GenAI solutions to Indian businesses.
© 2024 TechCrunch. All rights reserved. For personal use only.
AI startup Runway’s second annual AI Film Festival showcased movies that incorporated AI tech in some fashion, from backgrounds to animations.
© 2024 TechCrunch. All rights reserved. For personal use only.
AI startup Anthropic is changing its policies to allow minors to use its generative AI systems — in certain circumstances, at least. Announced in a post on the company’s official blog Friday, Anthropic will begin letting teens and preteens use third-party apps (but not its own apps, necessarily) powered by its AI models so long […]
© 2024 TechCrunch. All rights reserved. For personal use only.
In a post on Werner Vogels’ personal blog, he details Distill, an open-source app he built to transcribe and summarize conference calls.
© 2024 TechCrunch. All rights reserved. For personal use only.
Autonomous, AI-based players are coming to a gaming experience near you, and a new startup, Altera, is joining the fray to build this new guard of AI agents.
The company announced Wednesday that it raised $9 million in an oversubscribed seed round, co-led by First Spark Ventures (Eric Schmidt’s deep-tech fund) and Patron (the seed stage fund co-founded by Riot Games alums).
The funding follows Altera’s previous raising a pre-seed $2 million from Andreessen Horowitz and others in January of this year. Now, Altera wants to use the new capital to hire more scientists, engineers, and team members to help with product development and growth.
If the first wave of AI for end users was about AI bots; and more recently, AI “copilots” use generative AI to help understand and respond to increasingly sophisticated queries, then AI agents are emerging as the next stage of development. The focus is on how AI can be used to create increasingly more human-like, nuanced entities that can respond to and interact with actual humans.
One early use case for these agents has been gaming — specifically to use in games that support modifications (mods) like Minecraft. Voyager is one recent project, built on the Minedojo framework, that creates and develops Minecraft AI agents, and this, too, is where Altera is getting its start.
The company’s first product is an AI agent that can play Minecraft with you, “just like a friend” (the waitlist to try that out is here), but this seems to be just chapter one for the company. “We are building multi-agent worlds, opening up exciting opportunities in entertainment, market research, and more,” the company promises on its site. And after that? Robot dreams, it seems.
“Creating the human qualities required to turn co-pilots into co-workers and exploring a world where digital humans are given a physical form factor,” Altera explains.
At the helm of Altera is Robert Yang, a neuroscientist and former assistant professor at MIT. In December 2023, Yang and Altera’s other co-founders—Andrew Ahn, Nico Christie, and Shuying Luo—stepped away from their applied research lab at MIT to focus on a new goal: developing AI agents (or “AI buddies,” as Yang calls them) with “social-emotional intelligence” that can interact with players and make their own decisions in-game.
“It has been my life goal as a neuroscientist to go all the way and build a digital human being—redefining what we thought AI was capable of,” Yang told TechCrunch. That is not to say that Yang is coming from a misanthropic point of view. “Our solidly pro-human framework means that we are building agents that will enhance humanity, not replace it,” he insists.
What is notable about Yang and Altera’s focus is its consumer focus. This stands in contrast with a big swing that we have seen in AI towards building models that can be used either to speed up or sometimes replace humans in enterprise environments. (Even with OpenAI, ChatGPT has certainly been a viral hit globally, but at its heart the startup has been trying to build a business around usage of its APIs.)
“We see more potential in building agents within the gaming industry,” he said. “This approach allows us to iterate faster, collect data more effectively, and deliver a product where there are eager users and where emergent behavior is a feature, not a bug.”
(And yes, in keeping with its consumer focus, you should not be surprised that, for now, the company is not talking about monetization at all. )
Similar to the Voyager GPT-4-powered Minecraft bot, Altera’s autonomous agents are capable of playing Minecraft as if they were humans, performing tasks like building, crafting, farming, trading, mining, attacking, equipping items, chatting, and moving around.
Altera’s agents are designed to be companions for gamers, not assistants who do what you tell them to. Unlike NPCs (non-player characters), they have the freedom to make their own decisions, which could either make the game more entertaining or frustrating, depending on your playing style.
In a video demo, Yang plays around with multiple scenarios, including one where he tries to convince the AI agent to attack other people. The bot is hesitant at first, typing in the chat, “I don’t want any trouble, can we just find a peaceful solution? Fighting won’t solve anything.” Yang taunts it, commanding others to attack the “weak” bot. It eventually defends itself and kills Yang’s Minecraft character. “I’ll make sure they regret crossing me,” the AI agent wrote.
While the ending may be a little sinister, the gameplay feels no different from a regular session with friends, trolling and competing against each other.
Altera is currently testing the model with 750 Minecraft players and plans to officially launch later in the summer. It’ll be available via Altera’s desktop app, which is free to download but will also come with paid features.
Minecraft is just a starting point for Altera. The company eventually plans to bring the model to additional video games and other digital experiences. Altera’s AI agents “execute an action as code, meaning they can play any game without material customization,” Yang explained. For instance, it could work with Stardew Valley, he said. Altera will also integrate the technology with game engine SDKs for “broader developer use.”
In addition to the recent investments by First Spark and Patron, Altera has gained support from a long list of high-profile investors, demonstrating confidence in the company’s potential. Altera boasts investors such as Alumni Ventures, a16z SPEEDRUN, Benchmark partner Mitch Lasky, Duolingo Chief Business Officer Bob Meese, Vamos Ventures, Valorant co-founder Stephen Lim, and more.
“There exists a massive opportunity to create AI companions that engage in all areas of our lives. However, today’s AI lacks critical traits like empathy, embodiment, and personal goals, which prevent it from forming real, lasting connections with people,” Aaron Sisto, partner at First Spark Ventures, said in a statement. “Robert and the team at Altera are leveraging deep expertise in computational neuroscience and LLMs to build radically new types of AI agents that are fun, unique, and persistent across platforms. We are thrilled to be a part of their journey.”
Meta is rolling out an expanded set of generative AI tools for advertisers, after first announcing a set of AI features last October. Now, instead of only being able to create different backgrounds for a product image, advertisers can also request full image variations, which offer AI-inspired ideas for the overall photo, including riffs that update the photo’s subject or product being advertised.
In one example, Meta shows how an existing ad creative showing a cup of coffee sitting outdoors next to coffee beans could be modified to present the cup, from a different angle, in front of lush greenery and coffee beans, evoking imagery reminiscent of a coffee farm.
This may not be a big deal if the image is only mean to encourage someone to visit a local coffee shop. But if it was the coffee cup itself that was for sale, then the AI variations Meta offers could be versions of the product that didn’t exist in real life.
The feature could be abused by advertisers who wanted to dupe consumers into buying products that don’t actually exist.
Meta admits this is a possible use case, saying that an advertiser could tailor the generated output with the coming Text Prompt feature with different colors of their product, from different angles and in different scenarios. Currently, the “different colors” option could be used to dupe customers into thinking a product looked different than it does in real life.
As Meta’s example demonstrates, the coffee cup itself could be transformed into different colors, or could be shown from different angles, where each cup has its own distinct swirl of foaming milk mixed in with the hot beverage.
However, Meta claims that it has strong guardrails in place to prevent its system from generating inappropriate ad content or low-quality images. This includes “pre-guardrails” to filter out images that its gen AI models don’t support and “post-guardrails” that filter out generated text and image content that doesn’t meet its quality bar or that it deems inappropriate. Plus, Meta said it stress-tested the feature using its Llama image and full ads image generation model with both internal and external experts to try to find unexpected ways it could be used, then addressed any vulnerabilities found.
Meta says this feature has already begun to roll out, and in the months ahead, advertisers will be able to provide text prompts to tailor the image’s variations, too.
Image Credits: Meta
Plus, Meta will now allow advertisers to add text overlays on their AI-generated images with a dozen of the most popular font typefaces available to choose from.
Another feature, image expansion, also introduced in October 2023, will now be available to Reels in addition to the Feed, across both Facebook and Instagram. This option leverages AI to help advertisers adjust their image assets to fit across different aspect ratios, like Reels and Feed. The idea is that advertisers could spend less time repurposing their creative assets for different surfaces. Meta says text overlay will work along with image expansion, too.
One advertiser, smartphone case maker Casetify, said that using Meta’s GenAI Background Generation feature led to a 13% increase in return on its ad spend. The company had tested the option with its Advantage+ shopping campaigns, where the AI features first became available in the fall. The updated AI features will also be available through Ads Manager via Advantage+ creative, as before.
Image Credits: Meta
Beyond images, Meta’s AI can be used to generate alternate versions of the ad headline, in addition to the ad’s primary text, which was already supported by leveraging the original copy. Meta says it’s testing the ability for this text to also sound like the brand’s voice and tone, using previous campaigns as its reference material. Text generation capabilities will be moved to Mets’s next-gen LLM (large language model), Meta Llama 3.
All the generative AI features will become available globally to advertisers by the end of the year.
Outside of the AI updates, Meta also announced it would expand its subscription service, Meta Verified for businesses, to new markets including Argentina, Mexico, Chile, Peru, France, and Italy. The service began testing last year in Australia, New Zealand and Canada.
Now, Meta Verified will offer four different tiers to its subscription plan, all with the base features of a verified badge, account support, and impersonation monitoring. Higher tiers will include new tools like profile enhancements, tools for creating connections, and more ways to access customer support.
Meta Verified will be expanded to WhatsApp soon, the company also said.
OpenAI says it’s developing a tool to let creators better control how their content is used in AI.
Called Media Manager, the tool — once it’s released — will allow creators and content owners to identify their works to OpenAI and specify how they want those works to be included or excluded from AI research and training. The goal is to have the tool in place by 2025, OpenAI says, as the company works with creators, content owners and regulators toward a common standard.
“This will require cutting-edge machine learning research to build a first-ever tool of its kind to help us identify copyrighted text, images, audio and video across multiple sources and reflect creator preferences,” OpenAI writes in a blog post. “Over time, we plan to introduce additional choices and features.”
Thomas Li was working at Point72, the hedge fund founded by notorious investor Steve Cohen, when he realized that the financial industry relies heavily on manual data entry processes that could be prone to errors.
“As a buy-side analyst, I felt the pain of manually sourcing and entering data to build and update financial models,” Li told TechCrunch. “It took time away from the more important work of analyzing and making investments.”
After meeting Jeremy Huang, a former software engineer at Airbnb and Meta, and Daniel Chen, an ex-Microsoft engineer, through New York University connections (all three are all alums), Li decided to try his hand at an automated solution to the data entry challenges.
The three partners launched Daloopa, which uses AI to extract and organize data from financial reports and investor presentations for analysts. Daloopa on Tuesday announced that it raised $18 million in a Series B funding round led by Touring Capital, with participation from Morgan Stanley and Nexus Venture Partners.
“Daloopa is an AI-powered historical data infrastructure for analysts,” Li said. “This way of approaching the data discovery process keeps highly competitive firms and teams ahead of the curve.”
Daloopa’s customers are primarily hedge funds, private equity firms, mutual funds and corporate and investment banks, Li says. They use the startup’s tools to build workflows for investment and due diligence research. The workflows, powered by AI algorithms, discover and deliver data to analysts’ financial models, reducing the need to copy data manually.
“Daloopa provides a new way to get mission-critical data to both the buy side and sell side,” Li said. “The time savings is reinvested into research and analysis, or client-facing time — helping our customers gain an edge in their research process.”
Now, I’m a little skeptical that Daloopa’s AI doesn’t make mistakes: No AI system’s perfect, after all. Thanks to the phenomenon known as hallucination, it’s not uncommon for AI models to make up facts and figures when summarizing documents and files.
Li didn’t suggest that Daloopa is foolproof. But he did claim that the platform’s algorithms “only continue to improve over time” as they’re trained on growing sets of financial documents. Mum’s the word on where the data’s sourced from, exactly; Li says it’s a trade secret.
“Daloopa has been an AI company since birth five years ago, before all the AI hype,” Li said. “We’ve spent those years training our algorithms and developing AI for financial institutions.”
With the new funding, which brings NYC-based Daloopa’s total raised to $40 million, the company plans to grow its team of ~300 employees, bolster product R&D and expand its customer acquisition efforts.
“Daloopa is an AI-powered solution that started ahead of the curve and has seen year-over-year growth acceleration over the past two years,” he said. “As financial institutions increase their adoption of AI tools, we’re very well positioned to be a leader in the AI-driven fundamental data space.”
OpenAI is collaborating with Stack Overflow, the Q&A forum for software developers, to improve its generative AI models’ performance on programming-related tasks.
As a result of the partnership, announced Monday, OpenAI’s models, including models served through its ChatGPT chatbot platform, should get better over time at answering programming-related questions, the two companies say. At the same time, Stack Overflow will benefit from OpenAI’s expertise in developing new generative AI integrations on the Stack Overflow platform.
The first set of features will go live by the end of June.
The tie-up with OpenAI is a remarkable reversal for Stack Overflow, which initially banned responses from ChatGPT on its platform over fears of spammy responses.
Stack Overflow began experimenting with generative AI features last April, promising to craft models that “reward” devs who contribute knowledge to the platform. In July, the company launched a conversational search tool that lets users pose queries and receive answers based on Stack Overflow’s database of over 58 million questions and answers, along with tools for businesses to fine-tune searches on their own documentation and knowledge bases.
Some members of Stack Overflow’s developer community rebelled against the changes, pointing out concerns related to the validity of information generated by AI, information overload and data privacy for individual contributors on the platform.
There was at least some basis for those concerns. An analysis of more than 150 million lines of code committed to project repos over the past several years by GitClear found that generative AI dev tools are resulting in more mistaken code being pushed to codebases. Elsewhere, security researchers have warned that such tools can amplify existing bugs and security issues in software projects.
But despite the apparent flaws, developers are embracing generative AI tools for at least some coding tasks. In a Stack Overflow poll from June 2023, 44% of developers said that they use AI tools in their development process now while 26% plan to soon.
This has precipitated something of an existential crisis for Stack Overflow. Traffic to the platform has reportedly dipped significantly since the release of capable new generative AI models last year — models that in many cases were trained on data from Stack Overflow.
So now, as it cuts costs, Stack Overflow is pursuing licensing agreements with AI providers.
The company’s deal with OpenAI — the financial terms of which weren’t disclosed — comes after Stack Overflow partnered with Google to enrich Google’s Gemini models with Stack Overflow data and work with Google to bring more AI-powered features to its platform. Stack Overflow stressed at the time that the agreement wasn’t exclusive — and indeed, that turned out to be the case.
Prashanth Chandrasekar, CEO of Stack Overflow, previously said that 10% of the platform’s nearly 600 staff was focused on its AI strategy, and has described potential additional revenue from the strategy as key to ensuring Stack Overflow can keep attracting users and maintaining high-quality information.
“Stack Overflow is the world’s largest developer community,” Chandrasekar said in a press release this morning. “Through [our] industry-leading partnership with OpenAI, we strive to redefine the developer experience, fostering efficiency and collaboration through the power of community, best-in-class data, and AI experiences. Our goal with OverflowAPI, and our work to advance the era of socially responsible AI, is to set new standards with vetted, trusted, and accurate data that will be the foundation on which technology solutions are built and delivered to our user.”
Founded in 2017, Atoms Lanka Solutions is a provider of IT consulting and software development services. Throughout the company’s history, we have worked with over 160 customers from 5+ countries..
