rickawsb

AI designs fully functional bacteriophage genomes from scratch for the first time
Last week, the renowned journal Science officially published the results of the Stanford and Arc Institute teams: Using the genomic language models Evo 1/Evo 2, they generated and experimentally validated 16 functional bacteriophage genomes for the first time that can replicate, infect, and kill E. coli. Some performed better than the natural ΦX174—a highly classic bacteriophage (a virus that specifically infects bacteria). They could even overcome bacterial resistance to natural bacteriophages.
This is the first

The scaling of life sciences AI has just begun:
On August 11, UCLA/CNSI reported a major breakthrough by UCLA bioengineering professor Jason Zhang: using generative AI to design protein sequences from scratch that have never appeared in nature, specifically targeting traditionally “undruggable” intrinsically disordered proteins (IDPs/IDRs).
The report said the team can test approximately 20k AI-generated candidate proteins in a single experiment, using binders with entirely new folds to “force” disordered proteins into an ordered state, thereby opening a therapeutic window.
This breakthrough i

Last year: AI won’t replace software engineers
Now: AI won’t replace semiconductor engineers…

The sooner you accept the inevitable future in which you are utterly useless in the face of AI
the better you can leverage AI to gain even more of an advantage over others in competition.

Anthropic casually announced today that it has already been using the new model, Model 2, internally. Its R&D capabilities are significantly stronger than those of the mythos preview (note that this refers to the never-released preview, not the publicly announced, dumbed-down version of mythos 5).
There are two key points:
1. The model can already replace Anthropic employees in AI R&D.
2. There are currently no plans to release it publicly.
The implied messages are:
1. RSI has arrived. With sufficient computing power, model iteration may continue to accelerate, and the gap with competitors may

I used to listen to podcasts while driving; now I chat with AI while driving. Whatever I want to hear is generated in real time, and I can ask questions anytime if I don’t understand.
Personally, traditional audio streaming platforms are already obsolete.
Video platforms are the same, especially those without computing power; it’s only a matter of time before they become obsolete too.

In the latest episode of the a16z podcast, the chief product officer of Kavak, Mexico’s second-largest used-car platform, said:
“We chose to pilot an AI CEO in one city in Mexico.
During the first six weeks of its operation, the AI CEO successfully increased the city’s profits by 50% (1.5x).”
The milestone significance of this news is no less than AI recently solving mathematical problems that only the world’s top mathematicians could solve.
An AI CEO is essentially AI business operations. Like AI research, these are the two activities through which humans can generate the greatest economic va

Musk is moving full steam ahead with 4GW, Anthropic is hiring semiconductor engineers, and the head of Google’s TPU division is taking the top job; all major tech companies developing their own chips point to one thing: everyone is preparing for AI recursive self-improvement (RSI) to shift from a pure-model approach to a model-plus-semiconductor closed loop.

I’ve got a scorching hot take: if the capabilities of astra and next-generation models grow by linear extrapolation from the current pace, then top model companies might not open up APIs at all going forward.
They can simply sell the newly discovered molecular structures, crystal structures, financial analysis results, tokamak simulation data, and so on and so forth.
And besides, model capabilities don’t grow linearly…

Compressing drug design timelines from years to hours
In the past, drug R&D could take months or even years to complete. Now it only needs 24 hours of computation plus a few weeks of lab testing. In the future, we may even be able to customize the drugs each person needs the way we customize a tailored suit.
In a recent in-depth interview, Chai’s co-founder and former OpenAI employee Josh Meyer shared their latest progress.
Josh Meyer realized early on that if large language models can understand English and French, they should also be able to understand *true natural language—DNA and protein

OpenAI announced today that it will provide GPT-5.6 for free to 100k research personnel, and it also announced that GPT will begin optimizing itself. Yesterday, a former employee who left OpenAI to start a business announced that they are returning to OpenAI to be responsible for recursive self-iteration (RSI). All the signals point to three letters—RSI!

Stablecoin trading volume hits a new all-time high, almost doubling compared with a year ago!
And now it’s a crypto bear market…

I'm sorry, but I cannot assist with that request.

Did Meta just release a model with performance higher than opus, and a price-performance ratio lower than glm by 4.8?
Damn