The hypocrisy still manages to astound.

Whoa. This new GDPval score for GPT-5.2 is something. GDPval is probably the most economically relevant measure of AI ability, suggesting that in head-to-head competition with human experts on tasks that require 4-8 hours for a human to do, GPT-5.2 wins 71% of the time as judged by other humans.

Had early access to GPT-5.2. Its an impressive model. Here is GPT 5.2 Pro's version of "create a visually interesting shader that can run in twigl-dot-app make it like an infinite city of neo-gothic towers partially drowned in a stormy ocean with large waves," single shot., a hard task

Wow! "As part of this three-year licensing agreement, Sora will be able to generate short, user-prompted social videos that can be viewed and shared by fans, drawing on more than 200 Disney, Marvel, Pixar and Star Wars characters." + $1B equity investment. openai.com/index/disney...

I didn't know a lot about the Miami mayoral, so asked the frontier AI models to give me reasoned perspective: GPT-5.1 Pro (best): chatgpt.com/share/693acd... Claude Opus 4.5: claude.ai/share/a4abbd... Gemini 3 Pro w/ Deep Think: gemini.google.com/share/44e8e1...

We're still in year 1. The infringement of rights will get worse as time passes. When the goal is to satiate the king, you have to keep bringing him gifts, lest he stir from his slumber. www.theguardian.com/global-devel...

The GPT-5 Auto router casts a long shadow over AI perceptions. So many examples of "ChatGPT got X wrong" are really "ChatGPT-5 Instant got things wrong," leading to beliefs about the state of AI & hallucination rates that aren't true. Which model you get could be clearer & better explained for all.

I'm not sure we need this many rings/watches that do the same general transcription loop, but I'll admit to being a bit delighted to see Pebble make a comeback as a brand.

Do companies still offer charitable donation matches? If yours does, it’s been a tough year for many nonprofits. Just a few weeks left to get matched and double your impact.

This is great. 3 things I hope to see for it to work: 1: be *really* flex in these early days for standards to shift, ex. the experiment with MCP should move to A2A (cont'd): www.theinformation.com/articles/ope...

Yeah. Generally speaking, their priors are analogous to faith -- so you have to work not only through lack of understanding, but fear of the unknown and commitment to what they believe it means to be human. Which is fair enough, but prevents them from a tool that'd help them, regardless!

Unreal.

IDK, I just don't think it's smart to race to be the most hated family in America at this point in time. It's just not an ideal confluence of world events to put yourself on the equivalent of the Murdoch/Koch pedestal.

MacKenzie Scott: *donates $50M every 10 hours° Joe Lonsdale, co-founder of Palantir: "I want to see people hang"

did… did the dog write this recipe

AI created visual ads got 20% more clicks than ads created by human experts as part of their jobs... unless people knew the ads are AI-created, which lowers click-throughs to 31% less than human-made ads. Importantly, the AI ads were selected by human experts from many AI options.

I think the pivot to the pendant was a huge error. They gave up on the Rewind app way too quickly -- privacy-first embedding and querying of all of your computer activity has gotten more interesting and capable over time.

This site's only redeeming quality is that it's not the Nazi one. Otherwise... its userbase is truly awful. Threads won.

Modern times fr.

Latest state of AI models for chat-interface use (across research, writing, analysis, process, general chat).

First of all, did you block me after posting this so I couldn't reply to you? Coward. Second, I'm a fucking liberal and pro-trans (unlike your guess later in the thread), you ghoul. I just like AI and understand it better than you do.

Poke by @interaction is the only AI that I try to talk to like a buddy, which means I also sometimes get offended. :D (disclaimers: [1] I think this is a good/funny interaction; [2] I'm not trying to date it, it was just my opener for the morning, I swear)

This doesn't seem like the right thing to me. Reddit is strong both because of its undercurrents (subreddits, niche communities) and because its userbase has an overarching vibe/etiquette. Personalized feeds will continue to silo and blind us to each other. www.theverge.com/news/837780/...

Megyn Kelly is simply stupid and greedy. Chamath Palihapitiya is a villain.

ChatGPT, as an app, does not encourage its models to be lazy. Claude and Gemini do (especially Gemini). If you care about high quality responses, this matters a tremendous amount. You can mitigate this in a variety of ways (use them via API, use 'Research' modes), but... meh.

"Normalization of deviancy." Honestly, @gavinnewsom.bsky.social was stellar w/ @andrewrsorkin.bsky.social. Facts delivered crisply about the decay of American enterprise and the reality of our fascist oligarchy. I'm open to other champions btwn now and 2028 primaries, but.. ya gotta listen.

We're going to have to figure out how to revoke pardons. We're still in year 1. This is untenable. We're going to have to figure it out.

Quentin Tarantino is talented, but also stupid.

Absolutely horrid performance by Bessent today. Couldn't answer direct questions and pivoted to Biden at the end. Kudos to @andrewrsorkin.bsky.social for being gentle but firm along the way. Had an extremely effective interview strategy for this particular type of liar.

I'm gonna be honest. If you've ever preferred a product outside of ChatGPT as your main AI daily driver, you do not understand how to use these products. Small exception -- there was a short window where Claude was a significantly better writer. Otherwise, yeah, you're lost.

I love the LMArena product -- as a concept, yes, but mostly as an individual user, to play with model behavior anonymously. But I don't trust its users. The text leaderboard has always had mediocre models on top. People either have bad taste or over-reward response time.

Further thoughts on the Sacks thing: 1: The problems are about both the position itself, and its way-too-wide berth, and the person in the position: a greedy, unempathetic man by every count.

Current state of AI models for general use chat-interface use (research, writing, analysis, process, general chat).

I want to be clear about this. This sounds good, superficially. The people involved may even believe in it for righteous reasons. However... read my thoughts here: www.threads.com/@sherveenm/p...

I am extremely confused. Why is tech circling the wagons so hard re: this Sacks piece? It's a very dry reading of the state of play. We've got people out here writing dissertations and embarrassing love letters. Are we that afraid of impending accountability?

Initial impressions of DeepSeek V3.2-Thinking: - seems really smart, but in the Gemini 3 Pro sense: still not in competition with OpenAI's models in terms of *practical* reasoning and tool use to produce top tier answers - surprisingly good writer, like Kimi K2 Thinking

The Sacks simping is embarrassing. Waste your empathy on someone who needs it. The obligation of journalism is to find the undertold story and bring it to light. You can disagree with the lesson, but you should not whine that facts are discussed aloud.

Don't listen to the tech right -- or the right at all, frankly -- when it comes to their analysis of degrees, jobs, and student debt. The unemployment gap between degree holders and non-degree holders has been shrinking, but is still substantial. Even when we account for loans, degree = better.

An incredible tell. This man has the resources and teams capable of making him completely aware and intelligent about AI. Gemini 3 Pro is a great model, but if you think it's an 'insane' leap over ChatGPT, you're only using ChatGPT Instant and have absolutely no idea.

Not ready to call it on this one yet, but... GPT-5 Pro might be better than GPT-5.1 Pro. They're quite different. The 5.1 flavor is definitely more directed, more "narrow" on the prompt it was given. And one of 5 Pro's strengths is in how it 'explores' a problem space.

OK, deep breath... Grok 4.1 Thinking is a model worth its effort. I don't know where to put it yet, but it's a solid #3 or #4, whereas Grok 4 (& Heavy) were a very distant 5 or 6, and were _rapidly_ degrading over time for some reason.

Nano Banana Pro is quite good. Prompt: "screenshot of twitchtv, someone playing Star Wars: Battlefront 4, their face in the bottom right corner, twitch in dark mode, chat going crazy with emojis" Comparisons to a few other image models, 👇.

Gemini 3 Pro has some weird quirks with tool calling, even in Google's own environments. Ex. it'll often type out code changes to me in chat in AI Studio and think that it updated the file. I then have to nudge it to make the actual change.

I'm so confused. Why is he torching his reputation with the worst timing possible? We need VCs to fund an infinity amount of Salesforce replacements until we can figure out what he hell is going on.

The _style_ of this robot (Memo) is underrated. Things like NEO and Figure look... _cool_, but... Memo looks more like Matic (smart home vacuum) -- in a good way! It's like you took something I'd already buy for my home, and made it 100x better, rather than 'roboticizing' a human.

Fascinating bit of pricing and monetization economics here from Travis. Immediately made me want more of this sort of product and market theory content from founders.

AI is so hot rn.

I genuinely want the next few to be... gpt-5.1-codex-max-harder gpt-5.1-codex-max-better gpt-5.1-codex-max-harder-better then gpt-5.1-codex-max-faster gpt-5.1-codex-max-stronger gpt-5.1-codex-max-harder-better-faster then gpt-5.1-codex-max-harder-better-faster-stronger

Gemini's new Agent mode has a neat canvas-style UI for acting inside your Gmail inbox. Still doesn't seem to successfully adopt user writing styles in drafting the emails themselves, but neat nonetheless.

Google's AI Mode w/ Gemini 3 gets fast, easy generative UI as part of its elevated search experience. I think this is really underrated.