We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Shaan Puri: a podcast for an hour-long drive
"If you've got an hour drive this week, listen to this podcast," Puri wrote, linking the episode on YouTube. His post doesn't name the show's creator. He gives three reasons to listen:
- "The A-B-Z framework to stop overthinking"
- "Why your hometown might be holding you back"
- "The story of 'Iron John'"
Packy McCormick: watch the Tesla Roadster reveal
In Not Boring's Weekly Dose of Optimism #215, McCormick says the second-gen Roadster reveal at a SpaceX test site near Waco "was worth the wait" . He describes a 0–60 time of about one second. Using 10 nozzles fed by a 600-liter compressed-air tank, the car drove up an upside-down ramp, spun around and set itself back down, then hopped over other Tesla vehicles . He adds that Musk said production starts in 2028. His advice: "Just… watch the video if you haven't."
Satya Nadella: Microsoft Research on mid-range weather forecasts
Nadella pointed to his company's own work, published in Nature, so this is self-promotion. The research shows machine learning can improve forecasts two to six weeks out, a range that has been "historically been very difficult to predict." He cites uses in extreme-weather preparation, managing natural resources and farming decisions . His broader lesson: the future may depend on "intelligent verification systems" around models that "learn where models are systematically wrong" and keep checking predictions against reality. He links to Microsoft Research's write-up .
Shorter mentions
- Sam Altman shared Nick Cammarata's interpretability post: "Very happy to see this from Nick!" . The previous brief covered this post when Sarah Guo shared it. Cammarata says modern agentic tools make "the vast majority" of neurons legible in models up to trillions of parameters .
- Amjad Masad shared Haider's clip of Terence Tao. In it, Tao describes "the end of the romantic era of mathematics" and a move to an industrial era of large teams . Masad calls it "fascinating" to watch mathematicians go through the stages of grief, and says they are now "entering acceptance" .
- Aaron Levie shared an @MTSlive post. It reports that Anthropic's Michael Sellitto told the IMF that Claude could lead 80–90% of frontier model training by next year, up from 3% in April . Levie sees "an incredible flywheel" and says "If you aren't pricing in what RSI will lead to, that's probably a mistake." RSI is recursive self-improvement .
- Jason Lemkin promoted his own The Agents #016, "Going Headless: Claudeforce vs. Replit vs. Lovable." On the same data and prompts, the bills ranged from $2 to $15 (read) . Separately, he praised TBPN, saying Jordi Hays and John Coogan seem "back and in rare form" . He linked TBPN's clip of Jordi predicting a stunt-heavy Roadster launch .
- Chamath Palihapitiya promoted the new All-In episode with "Pod!!!!" . Ben Horowitz fills in for Friedberg. Topics include the CoreWeave debt selloff, AI's deflationary effect on software, Anthropic's IPO and Bill Gates's AI warnings .
Sam Altman praised Nick Cammarata’s X post on mechanistic interpretability, writing, “Very happy to see this from Nick!” and linking the post. Cammarata says modern agentic tools make the vast majority of neurons in models up to trillions of parameters legible, reversing his earlier pessimism, and predicts ambitious interpretability may be solved by the time it is most needed, possibly soon.
Packy McCormick recommends watching Tesla’s second-generation Roadster reveal video, highlighting its roughly one-second 0–60 run and stunt where the car drives up an upside-down ramp, spins, and lands; he concludes, “Just… watch the video if you haven’t.”
Chamath promoted an All-In Podcast episode with “Pod!!!!” and linked to it . The episode features Ben Horowitz filling in for Friedberg and covers the CoreWeave debt selloff, AI’s impact on software, Anthropic’s IPO, and Bill Gates’s AI warnings .
Chamath Palihapitiya endorsed a podcast episode with “Pod!!!” and linked the @theallinpod post; its announcement says Ben Horowitz fills in for Friedberg. Chamath gave no reason or takeaway for the recommendation.
Jason Lemkin (@jasonlk) shared a new video tied to this week’s The Agents, where he says they demo running Salesforce agentically in Claudeforce, Replit, and Lovable using real data . He claims these workflows are 10x–100x better than native Salesforce and says people may stop logging into Salesforce once Claudeforce is generally available . Video on YouTube
Jason Lemkin praised TBPN, saying Jordi and John Coogan were “back in rare form,” and linked TBPN’s video post featuring Jordi’s prediction that the Tesla Roadster launch would be a Nitro Circus-style demo with rocket thrusters, cars launching off ramps, and spins. TBPN post
Jason Lemkin promoted a new YouTube video as the latest installment of The Agents (watch). He says their rule for trying GTM AI agents is to do so only if 10K, their AI VP Revenue, wants to; he frames this as “Ask Your Developer” becoming “Ask Your Agent.”
Satya Nadella pointed readers to a Microsoft Research article about research showing machine learning can improve forecasts two to six weeks ahead; he highlighted applications in extreme-weather preparedness, natural-resource management, and agriculture, and the broader takeaway that models need verification systems to identify systematic errors, reason about cause and effect, and calibrate predictions against reality .
Jason Lemkin shared a new issue of The Agents, “Going Headless: Claudeforce vs. Replit vs. Lovable, Agents Picking Our Vendors, and Two of Our Agents Built Their Own Inbox to Talk Behind Our Back” (#016). The post highlights that the same data and prompts produced bills ranging from $2 to $15 and that agents now pick vendors. Read the issue.
Elon Musk amplified Arthur Macwaters’ X post, echoing its “Incentives drive outcomes” takeaway and linking to the post . Macwaters argues that paying NGOs according to the number of homeless people they manage will not reduce homelessness .
Aravind Srinivas promoted @morluto’s Reverse-Engineer-Anything (REA) skill, integrated into Perplexity Computer, and offered up to 10,000 credits for users to reverse-engineer an app of their choice; the linked post says the skill can inspect an app without its source code and build a working version (post).
Shaan Puri recommends this podcast for an hour-long drive, highlighting its segments on the A-B-Z framework for stopping overthinking, why one’s hometown might be holding them back, and the story of “Iron John.”
Aaron Levie quote-posted an @MTSlive post reporting Anthropic’s projection that Claude could lead 80–90% of frontier model training by next year, up from 3% in April; the post is linked here. Levie called the prospect of AI surpassing humans at model training remarkable and argued that stronger models, agent infrastructure, more data, and parallelized agents could create a flywheel accelerating model progress.
Clement Delangue praised Perplexity AI’s pplx-decider-v1.2 and pplx-decider-flash-v1.2 as “Beautiful work,” tagging Arav Srinivas and linking to both models on Hugging Face; he highlighted that their weights are available there: pplx-decider-flash-v1.2 and pplx-decider-v1.2. The release describes them as open-weight multimodal decision models, reporting improved text and vision scores, inference up to 5× faster, and Flash as the top 4B on the Decision Index.
Elon Musk endorsed @anabology’s “Make beautiful ads” video, made with @bot and @imagine, replying: “Yes! No more ads so ugly they make your eyes bleed.”
Amjad Masad amplified Haider1’s post quoting Terence Tao (with a video) and called the mathematicians’ response “fascinating.” Tao says mathematics is shifting from a romantic “age of heroes” to industrial, large-team work, creating new possibilities while some things are lost; Masad says mathematicians are moving from depression toward acceptance.
Martin Casado (@martin_casado) invites people to play with test scenes for the new Sparkjs for WebGPU at drone-space.netlify.app; he describes them as scenes to exercise Sparkjs. One showcased scene depicts space rays feeding off an interstellar cloud and uses about 400k splats, with transparency limiting the splat budget to help maintain usable FPS.
Aravind Srinivas shared “the physics behind the roadster,” linking to https://roadster.pplx.app/; the post gives no reason for sharing it.
