0:04
Anthropic launches its own drug discovery programs to tackle diseases Big Pharma considers unprofitable
Anthropic is launching its own drug development program for neglected diseases that the pharmaceutical industry considers unprofitable. Novartis CEO Vas Narasimhan thinks AI could cut development time from twelve years to seven or eight and double the success rate from 8 to 16 percent.
The article Anthropic launches its own drug discovery programs to tackle diseases Big Pharma considers unprofitable appeared first on The Decoder.
0:18
A 26,000-student study shows AI's hidden learning cost takes two full years to surface
A study of more than 26,000 Chinese students found that AI users finished homework faster and scored higher but performed up to 24 percent worse on exams. The full impact on entrance exam results took about two years to show up, meaning short-term studies systematically underestimate the damage.
The article A 26,000-student study shows AI's hidden learning cost takes two full years to surface appeared first on The Decoder.
0:32
OpenAI cofounder envisions "almost no interface" future where nobody learns software anymore
So I was reading about this idea that in the future, we might not need to learn software anymore. The cofounder of OpenAI, Greg Brockman, is thinking about a world where technology is so integrated that we don't need interfaces like we do now. He's talking about an agent that can understand context and just work without us having to do much of anything. It's pretty interesting because it's not just about making things easier, it's about changing how we interact with technology altogether.
What's notable is that this isn't just some pie-in-the-sky idea - it's based on the limitations of current technology. For example, OpenAI's own plugins for ChatGPT didn't work out as planned because the models just weren't advanced enough. So, instead of trying to force that approach, they're thinking about how to make the technology more seamless and invisible.
The key to this vision is creating an agent that can understand context and adapt to what we need. It's not just about having a smart assistant, it's about having a system that can anticipate and respond to our needs without us having to explicitly tell it what to do. It's a pretty different way of thinking about how we interact with technology, and it's going to be interesting to see how it develops.
1:07
Anthropic developer shares prompting tips for Fable 5 that focus on finding your own blind spots first
Claude’s Fable 5 feels less like a hardware upgrade and more like a reminder that the real bottleneck is often what we don’t see in ourselves. Thariq Shihipar says the model’s capacity is already there; the trick is surfacing the gaps we carry into the prompt.
He suggests a “blind‑spot pass,” where you deliberately ask Claude to expose assumptions you’ve baked into the problem statement—basically letting the model point out what you never thought to ask. It’s a quick loop: you write a prompt, Claude returns the hidden premises, you tweak, and you repeat.
Another tool is a structured interview with yourself. Write down the core question, then list the known constraints, then force a “what‑if” for each unknown. The exercise forces you to articulate the tacit knowledge that usually stays under the hood.
When those blind spots are cleared, Claude can take over the heavy lifting without tripping over the things you never voiced. It turns the model from a black box into a partner that helps you see the problem more clearly.
1:37
Stop Returning Text from RAG: The Typed Answer Contract That Prevents Hallucination
Enterprise Document Intelligence [Vol.1 #8A] - The schema is the contract: every field is a question the pipeline asks the model, and every answer is checkable
The post Stop Returning Text from RAG: The Typed Answer Contract That Prevents Hallucination appeared first on Towards Data Science.
1:47
Mistral's open-source Leanstral 1.5 aces formal math benchmarks and catches real bugs in code
Leanstral 1.5 isn’t just a bump in Mistral’s model line; under the hood it swaps the token‑mixing layer for a tighter, graph‑aware attention that lets it reason about Lean 4’s type system directly. That tweak lets the model breeze through the standard formal‑math suite, hitting top marks across the board. What’s cooler is that when the same engine scanned 57 open‑source projects, it flagged five subtle bugs that had slipped past human reviewers—issues ranging from off‑by‑one errors to mis‑typed API contracts. Those catches show the model’s practical edge: it can juggle abstract proofs and concrete code without losing its footing.
2:07
Microsoft follows Anthropic and OpenAI into the AI super app race with overhauled Copilot and AutoPilot agents
Microsoft reportedly plans to merge its consumer and enterprise Copilot apps into a single app in August. Rarely used features like Copilot Podcasts are getting cut, and new AI agents called "AutoPilot" will handle tasks in the background for an extra fee.
The article Microsoft follows Anthropic and OpenAI into the AI super app race with overhauled Copilot and AutoPilot agents appeared first on The Decoder.
