Models & Labs · United States

OpenAI’s math solutions aren’t meeting the field’s standards yet

Credited to TechCrunch · techcrunch.com · 3h ago

When OpenAI released hundreds of claimed solutions to some of the world’s hardest math problems this week, the frontier lab said that it had consulted an advisory group of elite mathematicians to avoid the controversy that came with the last time one of its models solved a long-standing problem in the field.

But OpenAI fell short of those standards, particularly where the mathematicians emphasized the need for human understanding of a mathematical result. That’s especially concerning after a new paper highlighted gaps between the natural language and formally expressed solution to a million-dollar problem ostensibly solved by OpenAI’s models.

The Advisory Group on Mathematics and Artificial Intelligence (AGMAI), hosted by Princeton University’s Institute for Advanced Studies, is made up of nine prominent researchers at institutions around the world.

The organization released guidelines for frontier labs solving math problems at the end of September. In a statement on the latest set of proofs, the AGMAI said that “it is ultimately up to the mathematical community to assess the extent to which our recommendations were followed successfully.”

However, the organization’s first request was “to stop testing advanced mathematical problems on proprietary models.” OpenAI’s release explicitly says that it is evaluating its proprietary models using open research problems in mathematics.

The advisory group did not respond when asked by TechCrunch for a more thorough evaluation of OpenAI’s latest proof release. The lab clearly followed some of its principles, including releasing results as soon as possible and including information about how the models reached their conclusions. But not for all of them: Just 10 of the 719 manuscripts included releases of the model’s chain of thought.

For papers that people don’t understand, the mathematicians suggested the proofs should be formalized — but just 42% of the proofs released by OpenAI had not undergone this process.

Ultimately, it’s still not clear that OpenAI is taking “responsibility for ensuring that human understanding will follow” when releasing its proofs, in accordance to the AGMAI principles. AGMAI suggested that OpenAI should help fund the work of human mathematicians who will be required to make the lab’s solutions meaningful in any real way.

“Problems are being solved autonomously by AI prompters who have no interest in the broader field itself once their initial target is ‘solved’, and do not understand the AI output well enough to answer questions on the result, give talks, or otherwise interact with the rest of the field,” Terence Tao, a prominent mathematician who has criticized OpenAI’s approach, wrote on social media after the release.

That problem is exemplified by a paper released this week by mathematicians at the University of Cambridge and King’s College in London that questions the way frontier labs are approaching these challenges.

When AI models solve mathematical problems, they first create a “natural language” explanation, then try to express that result in Lean, a programming language that in theory confirms the accuracy of the proof by compiling it as code.

However, there may be problems with the way the models translate their natural language proofs into code; this paper documents at least two discrepancies between the natural language proof and the Lean code behind the solution OpenAI has offered to a problem derived from the Navier-Stokes equations that describe the complex behavior of fluids.

These discrepancies don’t necessarily disprove either solution, but they do raise questions on whether we can simply rely on models to formalize their own solutions without human involvement. That’s one reason that AGMAI asked OpenAI to “include machine-readable metadata correlating the natural language and formal artifacts,” something that the frontier lab did not do with these releases.

“Because of the phenomenon of mistranslations — as highlighted in this paper — the NL proof by OpenAI and other autoformalised Lean proofs should not prima facie be trusted without the same peer review process and scrutiny that other proofs are subjected to,” the authors of the “lost in translation” paper conclude.

Mathematicians stress that when new results are discovered by humans, they take responsibility for them and engage with the broader community through papers, talks, and seminars. That process increases understanding of the solutions, finds strategies that can be used to solve other problems, and allows the new knowledge to be applied in practical fields.

When a model is prompted to solve a hard problem and spits out a solution, “there is not human understanding of them at the point of release, and now the work begins,” Harvard University mathematics professor Melanie Wood told TechCrunch.

Models & LabsUnited States

Google debuts SynthID Detector tool for flagging AI-generated content, but it’s far from perfect

Credited to SiliconANGLE · siliconangle.com19h ago

Useful

Google LLC is making its SynthID identification tools for artificial intelligence-created and AI-edited videos and images more widely available with the launch of a new service today that everyone can use.

The web-based SynthID Detector can identify images, video and audio clips that have been created using AI models from Google, OpenAI Group PBC, Nvidia Corp. and Kakao Corp. But in order to use it, people will have to sign in first, using a Google, OpenAI or Apple account. In a few weeks, the tool’s capabilities will be expanded to identify images created using Apple Inc.’s AI tools as well, such as Image Playground.

Google first developed SynthID in 2023. It’s an imperceptible watermark that’s added to AI-generated content created using tools from the above companies, but the only problem is that special tools have always been required to spot it.

Though the company has made different versions of its detection tool available to journalists and other media professionals, availability was severely restricted. Google also offers a few other ways to detect AI-generated content. For instance, there’s its Circle to Search smartphone feature, and users can also upload content to Gemini to ask if it has been created with AI or not. But with the launch of a dedicated detector tool, it’s making these capabilities more accessible,

The downside to SynthID Detector is that it has, well, downsides. For instance, because it’s designed to identify the hidden SynthID watermarks in AI content, it’s not very good at detecting AI-generated content that doesn’t have a SynthID watermark. And there’s a lot of that about.

Anthropic PBC’s Claude doesn’t use SynthID, and neither does xAI Corp.’s Grok chatbot. None of the many open-weights Chinese AI models use SynthID either. So it’s going to miss a lot of AI-generated content regardless.

Another issue with SynthID Detector is that it cannot distinguish between images, video and audio that has been entirely created from scratch or just edited using AI. So in the case of content that has been edited, it won’t be able to reveal which part is fake and which part is real.

SynthID Detector is still a step in the right direction, but it’s not nearly enough. The reality is that there’s still no foolproof way of verifying AI-generated content, partly because there are so many different standards for watermarking model’s creations.

Meta Platforms Inc. has been experimenting with its own watermarking tools and has developed a system for identifying images created with Muse Image, but it’s useless when confronted with images created with third-party models. Unless all of the major AI model makers, including the Chinese ones, decide to get on the same page and create some kind of universal standard for watermarking AI content, it’s likely that there will never be a reliable way to detect such material.

Given its influence and the fact it already has several technology heavyweights on board already, it’s likely that Google is in the best position to make some kind of universal standard happen. The company does at least seem to recognize the problem, and said on the SynthID Detector’s website that it’s “continuously working on extending our partnerships” to allow it to detect more types of AI-generated media.

“Building industry-wide standards takes time, and different content types (image, audio, video) have different technical requirements,” Google said. “The goal is a layered ecosystem where watermarking, metadata standards, and verification tools work together across the industry.”

Support our mission to keep content open and free by engaging with theCUBE community. Join theCUBE’s Alumni Trust Network, where technology leaders connect, share intelligence and create opportunities.

Are you an AWS customer? Support SiliconANGLE financially by buying your AWS services from our Marketplace portal page and links: https://siliconangle.com/aws-marketplace/

Founded by tech visionaries John Furrier and Dave Vellante, SiliconANGLE Media has built a dynamic ecosystem of industry-leading digital media brands that reach 15+ million elite tech professionals. Our new proprietary theCUBE AI Video Cloud is breaking ground in audience interaction, leveraging theCUBEai.com neural network to help technology companies make data-driven decisions and stay at the forefront of industry conversations.

Automation Anywhere acquires Boost.ai to expand customer-facing voice AI

Liquid AI builds personal AI around device-level context

Willow taps CoreWeave to simplify AI model training

Oratomic raises $475M for neutral-atom quantum computing, Universal Quantum nets $100M+

Docusign adds a desktop PDF editor to its Intelligent Agreement Management platform

Harness acquires Augment Code to bridge software engineering from idea to deployment

EMERGING TECH - BY KYT DOTSON. 4 HOURS AGO

Join our community

Original · SiliconANGLE

Front Method

Models & Labs · United States

With GPT-6, ChatGPT’s responses are much more visually appealing and interactive

Credited to SiliconANGLE · siliconangle.com · 21h ago

Photograph · SiliconANGLE

OpenAI Group PBC says ChatGPT is moving away from its traditional wall-of-text responses and giving users a more interactive user interface to play with.

In a media briefing today, the company said its GPT-6 models will now be able to access an “Intelligent UI” that will allow the chatbot to generate more visual outputs when it deems necessary, replacing words with graphics, tappable buttons, forms and charts” that will provide a much more interactive experience.

ChatGPT will determine for itself when an answer needs a more visually appealing response and interactive elements for the user to explore a topic more deeply. In such instances, the model will integrate these directly with its response.

In an example, OpenAI showed someone asking ChatGPT how to piece together a seven-speed bicycle. Rather than just churning out endless text that explains everything, it instead generated a complete design breakdown, complete with visualizations of each of the main parts and a clickable button that allows readers to learn more about each of them.

In a second example, a user asked ChatGPT to create a plan for a lamb roast with pictures of each dish, plus a recipe that allowed the user to alter the number of guests in order to recalculate the amount of ingredients required.

OpenAI explained that ChatGPT will generate these visualizations from a library of components that the GPT-6 models were pretrained on. For more basic prompts and questions, the chatbot will stick with the old text-only format, but users will have the option to request a more interactive response at any time, and it will reformat the response accordingly.

The goal with this update is to make “learning complex topics easier,” OpenAI Product Manager Aarush Selvan said in the briefing. The intelligence UI may also help to make ChatGPT more accessible and increase its appeal, though Selvan didn’t say this directly.

ChatGPT has predominantly been a text-based interface,” Selvan said. But the company wants to help people get things done, and it believes that “from the everyday things like planning a recipe or a trip or even trying to learn something, the most helpful answers aren’t just text.”

The visual UI is primarily designed to make ChatGPT easier for people to use and retain information, Holger Mueller of Constellation Research told SiliconANGLE. “We’ve long known that usability is one of the primary factors driving adoption of software, and the text-only interface doesn’t really help when people when it comes to information absorption or processing,” the analyst said.

Mueller believes that this is such a significant move by OpenAI that it could signal the opening of a new front in the battle of the AI assistants, especially if Anthropic responds with a similar visual aesthetic in Claude. “Anthropic will be under pressure to respond to this in the next few weeks, or it could even do so in just a few days if something similar was already in the works,” he said. “This could lead to usability and not just the quality of the model and its answers becoming the deciding factor in which chatbots people want to use. If Anthropic fails to respond to this, OpenAI has a very strong differentiator that would likely change many people’s choices once they see it.”

That is assuming that the Intelligent UI works as well as advertised, of course. It’s worth noting that it’s not an entirely original idea. In fact, it’s somewhat reminiscent of Google LLC’s “Generative Interface,” which was introduced with the launch of Gemini 3 last year, although it seems to be a bit more colorful and more graphics-heavy. When Google introduced that feature, there was a lot of excitement around it, but it doesn’t really appear to have differentiated Gemini as a chatbot all that much.

There may be other considerations, too. OpenAI didn’t say if the Intelligent UI would mean higher token consumption, but that does appear to be the case with all of those extra elements being generated, so that’s something users might want to watch out for. It’s also not clear how susceptible these visual elements may be to the problem that plagues every AI chatbot, namely hallucinations.

OpenAI, however, seems to be pushing for AI to become more visually appealing. Earlier this week, the company revealed it would soon introduce the first image-based ads in ChatGPT’s responses. More cynical users might think that the Intelligent UI may make it more difficult to see which responses are ads and which are the chatbots own. Regardless, ads are almost certainly going to become a part of every AI chatbot at some point.

OpenAI said the Intelligent UI is being made available from today for ChatGPT Plus, Pro, Business and Enterprise subscribers, though in the latter case, the workplace administrator will need to enable it. Free users and Go subscribers will get access to it tomorrow.