someone gave 20 llms the start of 668 obscure bacterial proteins (unknown function, no solved structure, 3 papers max per family) and asked for the next amino acid. then the same thing with each protein shuffled claude fable 5.2 got 54% on the real ones, 9% shuffled. opus 5.5 is level with protgpt3-10b, which is an actual protein model. gpt, gemini, deepseek, qwen, kimi all near zero so claude ate a lot of raw protein sequence in training. not papers about proteins, these ones barely have papers whole study cost about $730 in api calls. and anthropic's own filter blocked 44% of fable 5's questions as cyber content, for strings of amino acids
Sad CreatorShare
Source:Show original
Disclaimer: The information on this page may have been obtained from third parties and does not necessarily reflect the views or opinions of KuCoin. This content is provided for general informational purposes only, without any representation or warranty of any kind, nor shall it be construed as financial or investment advice. KuCoin shall not be liable for any errors or omissions, or for any outcomes resulting from the use of this information.
Investments in digital assets can be risky. Please carefully evaluate the risks of a product and your risk tolerance based on your own financial circumstances. For more information, please refer to our Terms of Use and Risk Disclosure.
