The biggest names in AI have converged on a single message: slow down. Rival leaders who normally compete on everything are now aligned on braking frontier development, and governments are being pulled in. At the same time, real-world AI misuse has moved from theory to documented fact, and new model designs are quietly rewriting the economics of running large systems.

Summary: Anthropic's Dario Amodei published an essay arguing frontier AI is moving too fast, in part because models can now improve themselves recursively. Sam Altman, Elon Musk and Microsoft's Satya Nadella backed the call, with Nadella saying building AI smarter than humans at nearly everything is not worth pursuing. Amodei proposed third-party evaluators inside labs, safety standards and coordination between nations. Critics, including former White House AI adviser David Sacks, dismissed it as virtue signalling and suggested looming IPO ambitions are the real driver.

Why it matters: When competitors this fierce align on caution, it signals either real fear about unreleased capabilities or a coordinated move to shape regulation in their favour.

Summary: Anthropic's threat report details real abuse of Claude, including a Yemen-based group using it for ballistic missile development and coordinated political disinformation across multiple countries. The company logged 35 suspicious biological weapons queries in 30 days, with five detailed cases involving dangerous pathogens and toxins. Some users rerouted traffic from unsupported regions and misled Claude about their intentions. Anthropic says it disrupted each operation and reported them to authorities.

Why it matters: This shifts AI risk from hypothetical superintelligence debates to documented misuse happening now.

Summary: DeepSeek's new V4.1-Flash is a 552 billion parameter model, nearly twice the size of its predecessor, yet uses a quarter of the memory cache. It scored 40 on one intelligence benchmark against 41 for a rival, at a quarter of the cost per task. The gains come from architectural changes, including a split encoder-decoder design and sliding-window attention, cutting cache from 48 KB per token to 890 bytes. It supports context windows up to one million tokens.

Why it matters: It breaks the assumption that more parameters means higher serving costs, pointing to design rather than size as the cost lever.

Summary: Google DeepMind and HHMI Janelia mapped the complete wiring diagram of a male fruit fly brain. Developers wired the connectome into projects by assigning functions to neuron signals, running it in Doom using damage as reinforcement, Beat Saber from replay data, and even Bitcoin trading by converting market data to neural inputs. Others tried Mario and a banana-finding robot. Researchers are exploring whether the map could seed new AI models, jokingly dubbed a fly-language model.

Why it is interesting: It is an early test of whether a real biological wiring diagram can serve as the basis for trainable AI.

Summary: Researchers used AI to analyse vocal patterns and semantic flow in speech, distinguishing people with schizophrenia from healthy controls with up to 87% accuracy. Americans with psychotic disorders wait an average of 18 months for a formal diagnosis, so faster screening could matter. The tool picks up subtle speech features humans miss. Experts caution it needs far more development before clinical use.

Why it is interesting: A voice recording could one day flag serious conditions long before symptoms are formally diagnosed.

Source: Smithsonian Magazine

Summary: A detailed multi-omic study of Maria Branyas, who lived past 117, found standard markers of advanced age in her immune system and chromosomes. Yet she also showed unusually low inflammation and protective genetic traits, giving parts of her body a younger biological age than expected. Researchers think this could explain why some elders resist age-related conditions like blood cancers. It is the most detailed biological assessment of a supercentenarian to date.

Why it is interesting: It hints that resilience to ageing may come from specific, measurable biological traits rather than luck.

Source: Cell Reports Medicine

Context: Use this to prepare for a salary negotiation by turning ChatGPT into a dedicated coach that builds a full strategy rather than generic advice.

Prompt: Act as my salary negotiation coach. I will give you my role, company, location, current offer package, key achievements, my leverage, and the minimum terms I would accept. Based on that, recommend my opening ask, a realistic target, and a clear walk-away point. Tell me which parts of the package to negotiate first, give concise responses to the likely employer objections, and write a short script I can use on the call. Do not tell me to bluff.

Example use case: A candidate weighing a job offer pastes in their details and gets an opening number, priorities, objection rebuttals and a ready-to-use call script.

Summary: Jensen Huang told a technology conference that Nvidia's revenue could jump 70% to around 680 billion dollars next year. He pointed to the company's near-monopoly on AI chip supply and claimed a 1 to 100 return ratio on its investments. Major labs and cloud providers depend heavily on its hardware.

Why it is important: It shows how much of the AI boom still runs through a single chip supplier.

Source: TechCrunch

Summary: OpenAI approached Congress to ask whether deliberately slowing AI development for safety reasons could violate antitrust laws. The question surfaces as rivals coordinate on caution. It frames safety as a possible legal grey area when competitors act together.

Why it is important: Coordinated slowdowns could collide with competition law, complicating any industry-wide pact.

Summary: King Charles plans to convene leaders from Nvidia, Google and Anthropic in Scotland to discuss safer AI and shared guidelines. OpenAI, Anthropic and Google are separately weighing a joint body for frontier AI testing and standards. Chinese state media framed the slowdown push as a strategy to hamper China's progress, and Bernie Sanders is calling for a global pause.

Why it is important: Governments and monarchs entering the debate could turn voluntary pledges into formal oversight.

That's it for today's Daily Pulse. Forward this to one person who wants to stay ahead of AI. See you in the next one. - Nicolas