Boris Cherny previews Anthropic's enterprise on-prem (1671 likes): 'We've been working on this with customers for a while. Mythos-class models require additional safety measures and enterprises need to meet their own privacy and compliance rules. Customers can own and control their own data and Anthropic retains none. It's coming this fall.' Thariq announces Fable safeguards for enterprises (543 likes): 'that work on your infrastructure so you have control over where your data lives & who has access to it. We've been developing it alongside ~100 companies for a bit and hoping to roll it out more broadly in the fall.'

Sottiaux's triple: the sub2api investigation (2797 likes): 'We've investigated a few messages about codex usage limits being different. What we did see is that when talking to affected users many were using sub2api. Converting a subscription into api traffic to then re-serve or share across many users is not something we support.' GPT-Image-2 transparency (2503 likes): 'Yay, you can now make transparent images in ChatGPT and through the API with GPT-Image-2. Here is a cactus cactus I am about to print and put on my laptop.' And ChatGPT Sites (1605 likes): 'Create a ChatGPT Site, share it and create something together. I've been building little games and sites with others and it is so much fun.'

Rauchg's triple: 'We're building AWS for agents' (387 likes); fx 0.0.5 gets even smaller (512 likes): 'fits in 2 floppy disks 💾💾 (xz-compressed). shipping tomorrow with our most-asked feature!'; and welcoming Bun to Vercel (463 likes): 'Bun's pursuit of simple, fast & open is the perfect match for Vercel and how we think about the world. Congrats @jarredsumner on shipping!' Amjad's origin story (938 likes): 'This partnership with @OpenAI is long overdue. Before Replit was in YC, PG asked Sam to recruit us. Here's Sam telling the story of how we first met.' Plus Free Mode speed notes (145/146 likes): 'One of the more underrated aspects of the new Free Mode is how fast it is. Making coding interactive again!'

Levie on post-training for applied AI (57 likes): companies closer to the underlying enterprise workflow will increasingly use post-training to bring down costs and improve accuracy — once you understand a domain well enough and have volume on a set of tasks. Madhu Guru's evals Part 4 (137 likes): a laddered eval strategy — hill-climb long evals pushing the production frontier plus multiple evals across the cost/realism spectrum. Nikunj's ambition math (112 likes): 'Anthropic became the fastest company to go from zero to a trillion dollars in valuation, beating every revenue record' — why ambition matters to investors.

Zara's Claude moment (295 likes): 'The other day I was feeling unmotivated and talked to Claude about it. Claude said something that completely changed how I see motivation: Motivation follows action more than it precedes it. Boom.' Peter Yang's manager-agent trick (484 likes): 'My gut is that a lot of AI output can be improved just by building a loop where a manager agent negs the other worker agent with prompts like: Are you sure this is the best you can do? I think you can do better, try again. Take a closer look, give me 11/10 output.' Plus his 100K YouTube subs (265 likes) while heading back to Vancouver to be with his mom.

Garry Tan's YC meme trilogy (210/400/537 likes): 'YC is the YC for AI Researchers', 'YC is the YC for consumer hardware', and an 'Interesting violation of you should dogfood your own product.' Swyx on the NVIDIA episode (60 likes): 'proud that Vibhu and I did the most recent pod on why NVIDIA just paid him $6B to buy the incredible model factory that is pumping out Thinky-beating models (actually not exaggeration, look at the numbers)' — plus Matt Pocock's /wayfinder (87 likes): 'when you are just navigating the fog of war and don't yet know what you don't know.' Aditya on SPC clarity (44 likes): 'One of the attributes we look for the most @spc is Clarity. The ability to be honest. To cut through the fog of war.' Plus 'The best founders are reductionists' (33 likes).

No Priors interviews Max Hodak, founder/CEO of Science (formerly of Neuralink), on Prima — the retinal prosthesis that just got European marketing approval and helps blind patients see again. Key threads: the brain very literally is a computer — the skull IS the brain-in-a-vat, connected through a small number of cables (cranial/spinal nerves); if you can get visual, auditory, balance, and motor signals in and out of the brain, that is an end in itself; the cochlear-implant-for-the-eye framing; why restoring vision beats drug discovery — 'you can do drug discovery for a decade, run a clinical trial, turn over a card, and the answer might be no'; the Prima chip under the retina with laser-projector glasses bypassing dead rods/cones; patients doing sudoku, crossword puzzles, reading books; the Pyxium acquisition nobody else saw value in; 'people forget that the moonshot worked — we left boot prints on the moon'; the platonic representation hypothesis — AI model internals look like neuroscience data, alignments between animal recordings and model representations; 'it feels like a law of physics: apply enough compute to matter and you get something that looks like intelligence'; the easiest way to do neuroscience today is ironically working on AI; BCI as a category like pharma (not one bet); silent-speech devices as hand substitutes — if you get vision, hearing, balance and a kilobit/sec of motor control, you're halfway to the Matrix; the border-redrawing question — communicating with a thing vs. integrating it as brain tissue; scan-copy thought experiments and why general anesthesia feels different; the brain as the only untransplantable organ — 'I'm going to be ultimately fairly disappointed if I'm murdered by my pancreas'; two leading causes of death (cardiovascular, brain-metastasized cancer) both attackable via this work; the 10-bits-per-second cognitive bottleneck — 'talking or writing IS thinking'; TAM: hundreds of thousands of patients now, millions with the next version in humans next year; $150K-per-patient (Second Sight) and ~$500K-per-eye (gene therapy) pricing precedents; and the long game — substrate independence, upgrading parts, fading the sense of jeopardy, and adapting ourselves to hard vacuum for the stars.