Claude 3.5

$new-secret-math-benchmark-stumps-ai-models-and-phds-alike$

New secret math benchmark stumps AI models and PhDs alike

AI, AI benchmarks, Anthropic, benchmarks, Biz & IT, Claude 3.5, Epoch AI, FrontierMath, Google, GPT-4o, large language models, machine learning, o1-preview, openai, research papers, vibemarks / 9u50fv / November 12, 2024

Epoch AI allowed Fields Medal winners Terence Tao and Timothy Gowers to review portions of the benchmark. “These are extremely challenging,” Tao said in feedback provided to Epoch. “I think that in the near term basically the only way to solve them, short of having a real domain expert in the area, is by a combination of a semi-expert like a graduate student in a related field, maybe paired with some combination of a modern AI and lots of other algebra packages.”

A chart showing AI model success on the FrontierMath problems, taken from Epoch AI's research paper. — A chart showing AI models’ limited success on the FrontierMath problems, taken from Epoch AI’s research paper. Credit: Epoch AI

To aid in the verification of correct answers during testing, the FrontierMath problems must have answers that can be automatically checked through computation, either as exact integers or mathematical objects. The designers made problems “guessproof” by requiring large numerical answers or complex mathematical solutions, with less than a 1 percent chance of correct random guesses.

Mathematician Evan Chen, writing on his blog, explained how he thinks that FrontierMath differs from traditional math competitions like the International Mathematical Olympiad (IMO). Problems in that competition typically require creative insight while avoiding complex implementation and specialized knowledge, he says. But for FrontierMath, “they keep the first requirement, but outright invert the second and third requirement,” Chen wrote.

While IMO problems avoid specialized knowledge and complex calculations, FrontierMath embraces them. “Because an AI system has vastly greater computational power, it’s actually possible to design problems with easily verifiable solutions using the same idea that IOI or Project Euler does—basically, ‘write a proof’ is replaced by ‘implement an algorithm in code,'” Chen explained.

The organization plans regular evaluations of AI models against the benchmark while expanding its problem set. They say they will release additional sample problems in the coming months to help the research community test their systems.

New secret math benchmark stumps AI models and PhDs alike Read More »

Claude AI to process secret government data through new Palantir deal

AI, AI ethics, Amazon, Anthropic, Biz & IT, chatgpt, chatgtp, Claude, Claude 3, Claude 3.5, confabulation, ethical ai, futurism, government contracts, large language models, LLaMA, machine learning, Meta, military AI, openai, palantir, US Defense Department, US intelligence agencies, Victor Tangermann / 9u50fv / November 8, 2024

An ethical minefield

Since its founders started Anthropic in 2021, the company has marketed itself as one that takes an ethics- and safety-focused approach to AI development. The company differentiates itself from competitors like OpenAI by adopting what it calls responsible development practices and self-imposed ethical constraints on its models, such as its “Constitutional AI” system.

As Futurism points out, this new defense partnership appears to conflict with Anthropic’s public “good guy” persona, and pro-AI pundits on social media are noticing. Frequent AI commentator Nabeel S. Qureshi wrote on X, “Imagine telling the safety-concerned, effective altruist founders of Anthropic in 2021 that a mere three years after founding the company, they’d be signing partnerships to deploy their ~AGI model straight to the military frontlines.“

Aside from the implications of working with defense and intelligence agencies, the deal connects Anthropic with Palantir, a controversial company which recently won a $480 million contract to develop an AI-powered target identification system called Maven Smart System for the US Army. Project Maven has sparked criticism within the tech sector over military applications of AI technology.

It’s worth noting that Anthropic’s terms of service do outline specific rules and limitations for government use. These terms permit activities like foreign intelligence analysis and identifying covert influence campaigns, while prohibiting uses such as disinformation, weapons development, censorship, and domestic surveillance. Government agencies that maintain regular communication with Anthropic about their use of Claude may receive broader permissions to use the AI models.

Even if Claude is never used to target a human or as part of a weapons system, other issues remain. While its Claude models are highly regarded in the AI community, they (like all LLMs) have the tendency to confabulate, potentially generating incorrect information in a way that is difficult to detect.

That’s a huge potential problem that could impact Claude’s effectiveness with secret government data, and that fact, along with the other associations, has Futurism’s Victor Tangermann worried. As he puts it, “It’s a disconcerting partnership that sets up the AI industry’s growing ties with the US military-industrial complex, a worrying trend that should raise all kinds of alarm bells given the tech’s many inherent flaws—and even more so when lives could be at stake.”

Claude AI to process secret government data through new Palantir deal Read More »

Anthropic’s Haiku 3.5 surprises experts with an “intelligence” price increase

AI, AI pricing, Anthropic, Anthropic Claude, API, Biz & IT, chatbots, chatgpt, chatgtp, Claude, Claude 3.5, Claude 3.5 Haiku, Claude Haiku, large language models, machine learning / Shannon Garcia / November 5, 2024

Speaking of Opus, Claude 3.5 Opus is nowhere to be seen, as AI researcher Simon Willison noted to Ars Technica in an interview. “All references to 3.5 Opus have vanished without a trace, and the price of 3.5 Haiku was increased the day it was released,” he said. “Claude 3.5 Haiku is significantly more expensive than both Gemini 1.5 Flash and GPT-4o mini—the excellent low-cost models from Anthropic’s competitors.”

Cheaper over time?

So far in the AI industry, newer versions of AI language models typically maintain similar or cheaper pricing to their predecessors. The company had initially indicated Claude 3.5 Haiku would cost the same as the previous version before announcing the higher rates.

“I was expecting this to be a complete replacement for their existing Claude 3 Haiku model, in the same way that Claude 3.5 Sonnet eclipsed the existing Claude 3 Sonnet while maintaining the same pricing,” Willison wrote on his blog. “Given that Anthropic claim that their new Haiku out-performs their older Claude 3 Opus, this price isn’t disappointing, but it’s a small surprise nonetheless.”

Claude 3.5 Haiku arrives with some trade-offs. While the model produces longer text outputs and contains more recent training data, it cannot analyze images like its predecessor. Alex Albert, who leads developer relations at Anthropic, wrote on X that the earlier version, Claude 3 Haiku, will remain available for users who need image processing capabilities and lower costs.

The new model is not yet available in the Claude.ai web interface or app. Instead, it runs on Anthropic’s API and third-party platforms, including AWS Bedrock. Anthropic markets the model for tasks like coding suggestions, data extraction and labeling, and content moderation, though, like any LLM, it can easily make stuff up confidently.

“Is it good enough to justify the extra spend? It’s going to be difficult to figure that out,” Willison told Ars. “Teams with robust automated evals against their use-cases will be in a good place to answer that question, but those remain rare.”

Anthropic’s Haiku 3.5 surprises experts with an “intelligence” price increase Read More »

Anthropic introduces Claude 3.5 Sonnet, matching GPT-4o on benchmarks

AI, Anthropic, Biz & IT, Claude, Claude 3, Claude 3.5, Claude 3.5 Sonnet, Claude Sonnet, Google Gemini, GPT-4, GPT-4o, large language models, LLMs, machine learning, Meta, Mistral, openai, text synthesis / Mike M. / June 21, 2024

Anthropic / Benj Edwards

On Thursday, Anthropic announced Claude 3.5 Sonnet, its latest AI language model and the first in a new series of “3.5” models that build upon Claude 3, launched in March. Claude 3.5 can compose text, analyze data, and write code. It features a 200,000 token context window and is available now on the Claude website and through an API. Anthropic also introduced Artifacts, a new feature in the Claude interface that shows related work documents in a dedicated window.

So far, people outside of Anthropic seem impressed. “This model is really, really good,” wrote independent AI researcher Simon Willison on X. “I think this is the new best overall model (and both faster and half the price of Opus, similar to the GPT-4 Turbo to GPT-4o jump).”

As we’ve written before, benchmarks for large language models (LLMs) are troublesome because they can be cherry-picked and often do not capture the feel and nuance of using a machine to generate outputs on almost any conceivable topic. But according to Anthropic, Claude 3.5 Sonnet matches or outperforms competitor models like GPT-4o and Gemini 1.5 Pro on certain benchmarks like MMLU (undergraduate level knowledge), GSM8K (grade school math), and HumanEval (coding).

Enlarge / Claude 3.5 Sonnet benchmarks provided by Anthropic.

If all that makes your eyes glaze over, that’s OK; it’s meaningful to researchers but mostly marketing to everyone else. A more useful performance metric comes from what we might call “vibemarks” (coined here first!) which are subjective, non-rigorous aggregate feelings measured by competitive usage on sites like LMSYS’s Chatbot Arena. The Claude 3.5 Sonnet model is currently under evaluation there, and it’s too soon to say how well it will fare.

Claude 3.5 Sonnet also outperforms Anthropic’s previous-best model (Claude 3 Opus) on benchmarks measuring “reasoning,” math skills, general knowledge, and coding abilities. For example, the model demonstrated strong performance in an internal coding evaluation, solving 64 percent of problems compared to 38 percent for Claude 3 Opus.

Claude 3.5 Sonnet is also a multimodal AI model that accepts visual input in the form of images, and the new model is reportedly excellent at a battery of visual comprehension tests.

Roughly speaking, the visual benchmarks mean that 3.5 Sonnet is better at pulling information from images than previous models. For example, you can show it a picture of a rabbit wearing a football helmet, and the model knows it’s a rabbit wearing a football helmet and can talk about it. That’s fun for tech demos, but the tech is still not accurate enough for applications of the tech where reliability is mission critical.

Anthropic introduces Claude 3.5 Sonnet, matching GPT-4o on benchmarks Read More »