pevidex

Launch party for my game. Guest list: me, my mom and a bot. The game starts the same way: an AI startup with no product and 3 users (you, your mom, a bot). Then you raise every round up to Series Z. Plenty of empty chairs. Free on iPhone, no ads: apps.apple.com/us/app/series…

I'm giving up proportional trial for my apps. I think I'll go back to the hard paywall, what do you think?

I'm selling, interesting call me.

Ai teaching in a video more than his classes in the classroom when he was 15 years old.

That's what I was expecting, tks @elonmusk

81 indie iOS apps on Plinth now. Nine days ago there were 44. It's free to submit, there's no way to pay to get in, and a person answers every one. Send yours. ontheplinth.com

If it's to make an bad MCP, better not even do @GoogleAds I'm expecting something innovative from you, don't disappoint me.

Anthropic says Claude formalized Fermat's Last Theorem in 30 hours, and I want to know who checked the 30 hours. Fable 5.2 and Mythos 5.2 are out, and the pricing and the access rules matter more than the proof. Claude Fable 5.2 and Claude Mythos 5.2 are the same underlying model. Fable has strong cybersecurity and biology safeguards, so it is open to all users. Mythos is for organizations doing beneficial cyber or biomedical work, and they apply through the Life Sciences and Cyber Verification Programs. During prerelease testing, the model designed three working electronic devices from scratch. It also solved two open problems in the physics of black holes and made substantial progress on robotics evaluations. Testers say it communicates more clearly than past Fable and Mythos models. Outputs come up to 30% faster than 5.1, and the price is $8 per million input tokens and $40 per million output tokens. There is also an optional fast mode, up to 2.5 times faster, at $16 input and $80 output per million tokens. Anthropic says Fable 5.2 at High effort is about level with Opus 5.5 at Max effort, for roughly half the cost per task. Their own advice is to start most coding and knowledge work on Opus 5.5, and move up to Fable or Mythos only when the work needs it. That reads to me like a price tag in disguise, and I think it is the honest part of the launch. If it can formalize a proof in 30 hours, what would you want it to build first?

The credit card line from Jefferies is the best way I've heard AI debt described. CoreWeave's 2032 bond with a 9.625% coupon now yields above 16%, and its default protection is around 1,155 basis points. $NVDA is roughly 139, so I'd read that gap before the headline.

Is a $529,344 $QCOM sale from one insider a signal, or just a person selling shares? I'd want the next trade by the same name before I bet on either.

Claude Fable 5.2 (Max) scores 65 on the Artificial Analysis Intelligence Index, and the median for comparable models is 40. The bill is $26.54 per task, which is last of 73 on cost. I'd pay for the smarts, but 45 minutes per task is a long wait for one answer.

HyperFrames from @HeyGen might be the best free repo for solo devs right now. Claude Code writes the video as HTML and @HyperFrames_ renders the MP4 on your machine. a launch video for your app without opening a video editor, I'm sold. github.com/heygen-com/hyperf…

Congress gets up to 45 days to report a trade. CEOs get 2 business days. If you want to follow smart money, the faster filing is also the one with research behind it: nitter.gq/pevidex/status/2119712…

Google's own help text says don't connect your apps if they hold anything you wouldn't want a reviewer to see. Read that as a self-taught person who ships apps alone and the sentence is strange. Who reads that before clicking connect?

Twenty words per instruction sounds strict until you read a wall of AI text. Then it sounds generous. Karpathy's 80% version keeps the short sentences and lets plain words back in. That is the part I would keep.

Every word of the PDF was written by hand, the repo says, with no AI writing. I use AI coding for all my apps, so I respect that line. The engine itself runs all 16 layers of Llama 3.2 1B on the GPU.

Why would a test runner live in a spreadsheet cell? recalc's answer is that you can see it when an agent is wrong about being done. Each cell can hold a number, a formula, a shell command or an AI agent. Agents fill the cells in real time, and you can open any one of them in its own terminal. The example I'd check first is the 10-K tie-out: sixteen figures from a public annual report, each with the line it came from. Formulas check that they agree, down to a $14.4 million difference the notes explain. Steering works two ways. You can prompt the agent, or you can edit the cells it fills instead of interrupting its chat. Your board stays on your computer, with no account, no server and no tracking. For private documents you can run the agents on a local model. Would you rather have the tests in the same grid as the agent, or in a separate terminal you have to go find?

Trusting my own code with my GitHub keys is already a stretch. Now a remote machine can borrow my local SSH login and push to my private repos. Friction turns out to be a feature. Who wants that on by default?

Anthropic says Fable 5.2 formalized Fermat's Last Theorem in 30 hours during prerelease testing. Same model as Mythos 5.2, which is gated for cyber and bio work. Fable is open to everyone at $8 in and $40 out per million tokens. Would you pay $16 in for the fast mode?