Anthropic is running a wet lab in San Francisco where its models do the experiments
Eric Kauderer-Abrams, head of life sciences, to Reuters: "We believe that to do biology, the final test is still, and will be for a while, in real lab work. We absolutely are doing that today"
Same week: a Life Sciences Verification Program that gives vetted researchers access to more capable models, and a drug discovery partnership with Novo Nordisk
It bought Coefficient Bio in April for $400 million
Anthropic would not say what the lab works on, only that it is not drug discovery - that would compete with its pharma customers
-
"not drug discovery" doing a lot of heavy lifting in a week that also had a drug discovery partnership in it
"Be fearful when others are greedy, and greedy when others are fearful"
People usually turn that line into a reason to buy on any red day.
For Buffett it worked the other way around: first a business he understood, then the price.
Berkshire could go years buying almost no public stocks and sitting in cash, if the market refused to offer the discount he wanted.
For ordinary people he advised not to play that game at all, but to hold 90% in a cheap S&P 500 fund and 10% in short-term Treasuries.
His own capital he built through Berkshire and a narrow set of names.
By mid-2026 there were about 29 public positions, with most of the value sitting in the top four or five.
On September 18 he became chairman emeritus.
The terminal shows the cash pile climbing to $397 billion by March 2026, and $134 billion of net selling in 2024 alone that is where sorting companies this way ends up.
A chatbot's bad answer nearly put a Chinese ship under fire
US Special Operations Command analysts fed open-source data and classified signals intelligence into an AI chatbot during the Iran war this spring
It misread the vessel's cargo manifest as nuclear weapons components
The analyst used the same tool to write that finding into an official summary, which moved up through command channels
Aircraft were already in the air when the operation was called off
CNN reported it Friday, citing sources. The model was not named
Over 15,000 repair companies on Google Maps. None of them existed
Every one had a local address and a local phone number. The addresses belonged to other businesses. The numbers rang overseas call centers
Premium Home Service, a Chicago company, built them for years. 7,600 phone numbers across 250 area codes, all aimed at the searches people make when something breaks: plumbing, heating, electrical, garage doors
The ratings did the selling. Fake five-star reviews buried the real one-star ones, under photos lifted from the staff pages of unrelated websites. One belonged to a professor at a North Carolina university
A caller paid a fee to what looked like a local firm. The job went to whatever contractor could be found. Some came late, some never came. One customer was charged $175 for 10 to 15 minutes on a floor heating thermostat. After the technician left, the basement wall was hot to the touch and smoking
The owner's version, according to the complaint, is that he knew "there were reviews created by some third parties which were not legitimately produced by customers"
The complaint says the reviews were posted by him, his employees and his relatives
Over 100,000 customers. Tens of millions of dollars. An FTC commissioner said it "defeated the efforts of diligent consumers"
Reading the reviews first didn't protect anyone. It's the exact habit this was built on
jev is an insanely cool product
Saw a post about jev and decided to check it out
This thing is straight up gold whoever built it is a genius
jev is the future
Scanned over 700 live ads in 40seconds flat, and it cost me just 4 cents.
If you're in marketing, this kind of crazy fast data crunching is hands down the best thing out there.
Doing this through Opus 5 would run around 2 to 4.5 million tokens, set you back $15 to $50, and take anywhere from 10 minutes to an hour. That’s 200 to 500 times more expensive than Jev, and way slower.
Gemini 4 Pro benchmarks compared to other models
Actually, no benchmarks have been officially announced for this version yet
The ones people are sharing right now are just from older Gemini models
OpenAI, January 16: "We keep your conversations with ChatGPT private from advertisers"
OpenAI, August 31: "Advertisers do not receive access to people's private conversations"
OpenAI, September 16: a new ad where you open a chat with a business's own agent. The post calls that chat "distinct from ChatGPT's independent answers and separate from the original conversation"
The promises cover your private conversations with ChatGPT. This one is a conversation with the advertiser, and none of the 3 posts says what the advertiser gets out of it
Сan't leak ur chats to advertisers if the advertiser is the one ur chatting with
40 hours, 100 teams, unsolved problems in research mathematics, and over $2M in credits from OpenAI and Anthropic. DARPA's expMath programme and Cognition are also listed as backers
The prize for the strongest solutions in round one: more credits
Asked about actual money, one of the organisers wrote that "there are a lot of regulations around paying people cash. We are doing the paperwork. If approved, we will give out cash prizes to participants and honoraria to judges"
If approved
771 mathematicians signed an open letter against the event. Their sentence about the students: "This labor goes uncompensated, uncredited, and unacknowledged." They also wrote that AI companies "see research mathematics as an advertising opportunity" and "will take credit for the effort of talented undergrads"
OpenAI pulled out. Its research lead said the rapid progress of AI in mathematics "is disruptive" and that the company wants to engage with the math community more
Strip the framing and it's a weekend of unpaid research on somebody else's model, with the compute donated by the two firms whose models need the problems
The sponsors paid in the only currency that comes back to them
1,000 married Americans were surveyed, and this is the number: 33% said AI tools "get" their relationship struggles better than their spouse does
44% have used AI for relationship advice. Among millennials it's 65%
64% have searched for marriage help online before talking to the person they married. They ranked digital sources above friends and family at 30%, and above a therapist at 22%
30% then took a real-world action off that advice. 28% made a financial decision on it without telling their spouse, and the groups most likely to do that were men and people earning over $125,000
44% said the AI left them calmer. 41% said it left them feeling understood
Hold those last two next to each other. Calmer, understood, and a decision the other person doesn't know about yet
The chatbot has only ever heard your side. It has never met them, never sat through the version where they're right, and it doesn't have to live in the house afterwards
Calmer and understood is just what it feels like when nobody in the room disagrees with u
People keep misunderstanding this. I CONTRIBUTED to Bostrom’s Superintelligence book and he thanks me by name in the foreword.
I was thinking about AI safety long before 2014.