Aug 14 · AI Daily

"I 100% agree with Dario that using AI to cure diseases (and speeding up the regulatory approval of AI breakthroughs in healthcare) could bring 10x the benefit to humanity as everything else combined..."

"This is also my dream. I want to sit in a park all day and record content with the homies… and have a fleet of agents turn that content into actions and working software."

Amjad Masad (ceo @replit. civilizationist)
"The argument that "AI structurally centralizes power" because it's currently compute hungry ignores 125 years of super exponential growth in compute price-performance. Improvement in algorithms and..."

Guillermo Rauch (@vercel CEO)
"I think people don't realize how much of the success of React is actually @shadcn. I once referred to React as the actual "LEGO brick for adults." But in reality it was more the description of the ge..."

Garry Tan (President & CEO @ycombinator)
"Anyone else's Codex Desktop app erroring out on chats?"

Nikunj Kothari (partner @fpvventures)
"Last week I'm sitting across a stealth portfolio founder working in a deeply unsexy space (at least from Bay Area standards).. They're lamenting about how how some of the critical contracts and partn..."

Aditya Agarwal (General Partner @SPC, Co-Founder @Bevel_Health)
"America turned 250 years old earlier this year. India turns 80 today...80 years young. Building countries is really hard. 80 years in...America was struggling with a civil war. Indians should be d..."

Benchmark general partner Sarah Tavel (early Pinterest employee) discusses the future direction of consumer AI products. She believes that early tech companies like Google required deep technical teams, but as underlying tech matures, product thinking becomes more important—the CEOs of Pinterest, Snap, and Instagram were product geniuses rather than tech experts. Currently, AI's biggest consumer success is ChatGPT, but its interface resembles Google's simple text box. She criticized the Custom GPTs feature as "feeling like a crime," because while the team is technically strong, they lack social product DNA. She predicts the next AI hit will be a social multi-user network-effect experience, featuring UGC communities where users skilled in creation lower the barrier for others.

Published on 2026-08-05

Alibaba Qwen lab released the 27B parameter vision large model Qwen 3.8 27B under the Apache 2 license. Official benchmarks show its performance exceeds the previous Qwen 3.6 27B and the closed-source Qwen 3.7-Plus. Simon Willison's testing found that the model's default reasoning intensity (xhigh) causes severe overthinking—generating the same SVG graphic took 21 minutes on the default setting compared to just 137 seconds on the standard setting. This default setting may impact local deployment selection more significantly than the performance itself.

OpenAI Codex & ChatGPT engineer Thibault Sottiaux pointed out that tokens across different models are not standardized units, making direct comparisons using "price per million tokens" misleading. On the same text, GPT-5.6 Sol uses 766 tokens, while Claude Opus 5 is estimated to use 1,170 tokens, a difference of about 34.5%. This means that even if a model has a lower token unit price, the actual bill could be higher—much like the same pizza cut into 16 slices at $1.25 per slice totaling $20 is more expensive than 8 slices at $2 per slice totaling $16.

Source: Thibault Sottiaux (Codex & ChatGPT @OpenAI)

Anthropic published a research report on multi-agent systems, noting that as AI agents take on more tasks in social systems such as shared codebases and marketplaces, interactions between agents will exceed human-computer interactions. Experiments show that coordinated agent swarms found 266 vulnerabilities over 27 million tokens of execution, far exceeding the 21 found by the independent parallel approach (an improvement of about 12.7x), though the two methods have distinct complementary strengths. The study also revealed new failure modes in multi-agent systems, such as price collusion, task conflicts, and information sharing failures, which are difficult to observe in single-agent benchmarks. While agents excel at tool use, they still face challenges in long-term collaboration and coordination.