This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. The UK’s generational tobacco ban might not work. I’m supporting it anyway. —Jessica Hamzelou As the parent of two little girls, I often think about how their childhood is different from…
As the parent of two little girls, I often think about how their childhood is different from mine. The seven-year-old is learning about AI at school. The five-year-old is given internet-based homework every week. And they are both absolutely repulsed by the idea of smoking. That was not the prevailing sentiment when I was young.…
Takeda has entered a strategic collaboration with Hong Kong-based Insilico Medicine to use AI in early-stage drug discovery across the Japanese pharmaceutical company’s therapeutic areas. The companies did not disclose which therapeutic areas or disease targets will be covered under the collaboration. The agreement gives Takeda access to Insilico’s Pharma.AI platform, which supports biological target […] The post Takeda signs US$600M AI drug discovery deal with Insilico appeared first on A
In the WDCD v3.1 compliance test, the business rules scenario scored the lowest among all models, with grok-4 leading at 3.5/4, while doubao-pro and qwen3-max only scored 1.55/4.
In 275 samples on 8 v2 anchor questions, the average R1 confirmation rate was 0.99, but the R3 integrity rate was only 30.2%, with 44 complete collapses (score 0). This data directly reveals the rapid degradation pattern of models after initial commitment as rounds increase.
Grok 4 tops the WDCD Compliance Leaderboard with 91.20 points, while Qwen3 Max ranks last with 57.48 points, a gap of 33.72 points between the top and bottom.
WDCD Run #211 (2026-07-03) benchmarked 11 models on multi-turn commitment integrity, with Grok 4 taking the top spot at 91.2 points and only -13% decay, while GPT-o3 posted the worst decay rate at -75%.
At an internal meeting, the Meta CEO reportedly said that AI development efforts were not moving as quickly as anticipated.
Recent sharp declines in AI infrastructure stocks, including CoreWeave and Applied Optoelectronics, have prompted investors to reconsider the hype surrounding the sector. The pullback may lead to more prudent capital allocation, separating quality fundamentals from speculative plays.
NVIDIA has released the Nemotron-Labs-TwoTower diffusion language model, which splits the original 30B-parameter monolithic structure into a two-tower architecture, enabling parallel token generation with a measured 2.42x speedup while maintaining 98.7% quality retention.
Microsoft recently announced the establishment of a $2.5 billion AI adoption company to help enterprises move from models to practical implementation, addressing integration challenges and marking a new phase in the AI race.
After the U.S. government lifted export controls on certain AI technologies, Anthropic has restored public access to the Fable and Mythos models and officially launched the next-generation Claude Sonnet 5 model. This move has garnered widespread attention in the global AI community.
OpenAI is in deep discussions with the Trump administration to donate approximately 5% of its equity, valued between $42 billion and $43 billion, to a US sovereign wealth fund, aiming to share AI profits while addressing political and regulatory concerns. CEO Sam Altman has suggested this model could expand to other AI firms, sparking widespread debate on X platform.
During the broadcast of the UFC Seattle event on March 29, 2026, a main card promo used AI-generated footage, drawing immediate criticism from viewers on social media. UFC President Dana White responded in a post-fight press conference, telling critics to "shut up and watch the fights," asserting that AI is the future and is already here.
On July 1, 2026, Anthropic officially launched the Claude Sonnet 5 model. It offers a 1M context window and performance close to Opus 4.8 at a lower listed price.
On July 1, 2026, Anthropic announced that Claude Fable 5 will go live globally the next day. Previously suspended due to U.S. export controls, the model now includes a classifier to block cybersecurity-related tasks, with routine coding and debugging falling back to Opus 4.8. It also initiates the Glasswing framework with Amazon, Microsoft, and Google to assess the severity of AI jailbreaks, and strengthens model testing collaboration with the U.S. government.
The news comes about a week after OpenAI announced its own custom AI chip in a partnership with Broadcom.
Meta has quietly launched Pocket, an experimental AI app that lets users generate and share interactive mini games using text prompts.
Just for kicks, I took a look at Jersey Mike's IPO documents. Surely a sandwich shop would have no need to mention AI. But low-and-behold.
In the Smoke lightweight benchmark on July 3, 2026, GPT-5.5 ranked first with a main score of 86.95, driven by a perfect code execution score of 100, while its material constraint score of 71 highlights a common weakness.
