OpenAI announces ChatGPT visual ads and expanded measurement

#1 📝 OpenAI News

Building advertising for the way people use AI - OpenAI announced on October 5, 2026, a new visual ad format that helps people explore products and services, alongside expanded measurement to help businesses understand ad impact. The post also invites businesses to start advertising in ChatGPT.

#2 ▶️

How the OpenAI team uses ChatGPT Sites daily

How I AI Podcast

#3 𝕏

A one-shot Pokémon FireRed Elite Four and Champion run was demonstrated using decision-making powered by Perplexity’s Decisions API. Across 137 live API calls, median response time was 592 ms, p95 was 987 ms, 96.4% of responses were under one second, and estimated inference cost was $0.028.

#4 ▶️

OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux

Lennys Podcast

Also covered by: @Lenny Rachitsky

#5 𝕏

Alexandr Wang recapped items described as shipped, including Muse, connectors, invite codes, a phone call beta, Muse for Mac, Muse in Canada, a connector platform, Mac computer use, Muse for small business, gadgets, memes, and “no butthole.” The post asked what should be shipped next.

#6 𝕏

Harrison Chase shared that coding-agent costs fell significantly for a second month in a row after three steps: tracking usage in LangSmith, setting user-level cost caps through an LLM gateway, and optimizing the OpenSWE harness with techniques such as model routing.

#7 ▶️

How to Use AI to Survive (and Even Enjoy) Meetings | Sam Stephenson

Peter Yang

#8 in

Peter Yang recapped Granola co-founder Sam’s comments that, although Granola initially tried to keep users in its UI, enterprise compatibility with internal tooling is essential. Granola is now comfortable with workflows that capture context in Granola and use it through an internal agent via the MCP.

#9 𝕏

DeepLearning.AI shared that, according to Anthropic, GLM-5.3 solves 12% of ExploitBench exploit tasks versus Claude Mythos’s 14%, while $20.40 in tokens found a recent Google Chrome security exploit. In this week’s letter, Andrew Ng argues these cybersecurity risks are an engineering problem and that defenders hold the long-term edge.

#10 𝕏

Boris Cherny said there were no changes on their end and usage is monitored carefully. He recommended running `/usage` for a token-by-token breakdown, noting that excessive usage is most often caused by a runaway loop or inefficient skill.

#11 𝕏

Aravind Srinivas shared that custom vertical AI apps can be built inside Computer, such as one that guesses an image’s spatial location using 3D and satellite views.

#12 𝕏

Udi Menkes 🚢 commented that every product now has “2 customers,” making his long-used brand model (becoming the name a person reaches for first) incomplete.

Also covered by: @Udi Menkes 🚢, @Udi Menkes

#13 𝕏

Guillermo Rauch said fx.sh is now much faster as harness overhead becomes increasingly important with faster models such as Astra ultrafast. He announced that the next release will further improve session storage and retrieval performance and deliver a major cloud-durability improvement in libfx.

#14 in

Guillermo Rauch predicted that security will become an increasingly large function in software companies, spanning verification engineering and capital allocation. He described increasingly sophisticated AI adversaries and distrust of a “2-person-and-a-dog company” as a challenge for startups; conversely, he said poor global cybersecurity and growing reliance on centralization create opportunities in areas that small teams can disrupt, which he is interested in investing in.

#15 𝕏

Teresa Torres shared an episode in which Steve of Vistaly says teams rarely change their minds until they hear customers describe problems in their own words. Vistaly puts interview evidence directly on opportunity solution trees to help teams hear, see, and align on the real problems.

#16 𝕏

Yann LeCun commented that while training frontier models is expensive, distilling them is cheap, a market force he argues indicates frontier foundation models will be free/open. He linked to thealliance.ai’s Tapestry project.

#17 𝕏

Boris Cherny commented that the spinner was not designed for Opus becoming this capable at long-running work and suggested adding “8d” to it.

#18 𝕏

Lenny Rachitsky commented “PM ascendant” on a post about UX work for personal agents (including instinct, dot, muse, and grok bot) that emphasized product touches over model upgrades.