Christopher Roebuck

@sama Well... I mean God, Family, Country/AI.

@sama Stay safe. Family first. AI second.

I wrote this early this morning and I wasn't sure if I would actually publish it, but here it is: blog.samaltman.com/2279512

@lindayaX @EliLillyandCo Approval is the starting line, not the finish. The real economics of oral GLP-1s depend on whether the $900+ WAC holds against the PBM-driven rebate wall. If pricing doesn't drop to reflect the lower manufacturing cost of small molecules, access stays throttled by PAs.

A big step forward. FDA approval of @EliLIllyandCo oral GLP-1, Foundayo, unlocks access for many more in need.

@BGOV GLP1s are the stress test for Medicaid budgets. High volume+long-term offsets = immediate fiscal crisis for states. We're ignoring 10-year savings on comorbidities because the budget clock resets every October. That's a structural failure, not a drug pricing one. @NIHDirector_Jay

Anti-obesity GLP-1 drugs are one of the hottest medications on the market in America right now. But with their increasing popularity comes issues with access, cost, and quality. news.bgov.com/bloomberg-gove…

@DeryaTR_ Derya, I forgot what the stakes were, wasn't it a lifetime subscription to Spud? @GaryMarcus pay up buddy!

Late 2024, I had a bet with Gary Marcus: if OpenAI's valuation stayed under $87 billion at the end of 2025, he would win; if it was over that, I would win. Obviously I won, except that the valuation ended up being about 10x more!😅Big congratulations to OpenAI! Much more to come!

@iruletheworldmo @MatthewBerman To be fair to @ManusAI , if it weren't for them-- there probably wouldn't be an @openclaw. It was the first really useful agentic system that got shit done. I was proud to have been in the first batch of users. @peakji is the man!

@MatthewBerman are they doing the thing?

They’re all building OpenClaw

I am starting to think that we are all just a collection of Markdown files.

@AlexFinn Alex is the most hyper-hyped AI guy I watch-- AND--I completely agree with him. I would only mention that "underclass" will be far better than most classes today--as AI drives prices toward zero. Still I would rather be nearer the front of the line. To the moon!

Drop what you're doing and read this post Anthropic is about to release the most powerful AI model ever created. Claude Mythos It's so powerful they consider it a danger to cybersecurity everywhere It will also be significantly more expensive than Opus I don't know your financial situation, but I do know this: we are entering the era of humanity where having money is going to give you signficantly more power than those without money Those without money will not be able to create nearly the same economic value as those that already have money because they won't have access to this kind of super intelligence, widening the gaps even further This is where the concept of a permanent underclass comes from If you want to keep up in this economic race you have to be willing to do 1 of 2 things: 1. Spend the money necessary to use this intelligence to create value 2. Procure the money you need to use this intelligence to create value The people who go out and immediately use this model to create business, products, and services will experience unmatched wealth Both the scariest and most exciting time ever

@kimmonismus I have invested many hours building my @OpenClaw, so I'm worried for where it's heading, but of course everyone's setup can easily be ported--as the sauce is in the markdown files and the config.json. But, indeed "open models aren't there yet." @steipete

Two possibilities: Either he's saying this now as an employee of OpenAI out of loyalty (which I find hard to imagine), or option two, because he's seen how far along the models are internally (more likely). Because his two statements strongly contradict each other.

Trying to be helpful with my #Clawdit skill. Your @openclaw agents cannot improve themselves without the bias introduced by their own system files. Same reason we need counselors and life coaches. I can't get the word out with 600 followers. But I tried. EOM. @steipete @MatthewBerman @AlexFinn @WesRoth

@kimmonismus I wish for impressions. Be thankful.

I've seen my impressions drop by about 60% for 1.5 months. Do other Premium+ subscriber have the same problem? is it a known bug?

@iruletheworldmo Sorry but nah, not with Semantic Scholar and arXiv. That doesn't adequately cover many disciplines. But what do I know.

this will be huge in a few days. and you heard it here first. from the strawberry man agi is coming.

7/ Open source. MIT. Built by a health economist who thinks about systems. GitHub: github.com/aceclaw826/cla… Try it on your own agent. The first audit is always revealing. @jordymaui @MatthewBerman @kimmonismus

3/ This is why therapy works. Not because a therapist is smarter than you — but because they're unencumbered. They don't carry your assumptions. They can see the contradiction between what you say you want and what you're actually doing. That separation is the entire mechanism.

1/ You'd never ask someone to be their own therapist. "Just think really hard about your blind spots." That's absurd. The mind doing the evaluation is the same mind holding the biases. You need an external, unencumbered observer. Your AI agent has the same problem. 🧵 @openclaw @steipete @MatthewBerman

6/ Contradictions between files. Stale model references. Vague directives. Redundancies burning tokens. Your agent can't see any of this. An unencumbered observer can. @coreyganim @WesRoth @openclaw

5/ It interviews YOU about your goals first — before reading any files. Your goals are the rubric. Then it reads every loaded file, breaks them into individual beliefs, and evaluates each against what you actually want. @austingriffith

4/ I built Clawdit — a separate agent with its own clean workspace. When it reads your SOUL.md, that file enters as DATA to analyze, not instructions to follow. The auditor's mind is clean. Non-negotiable. @cathrynlavery

@DutchRojas The structural problem: 340B has zero transparency requirements on how covered entities use the spread. HRSA audits <200 of 50,000+ entities/year. You can't fix misaligned incentives you can't measure.

The 340B drug pricing program was designed to help safety-net hospitals serve the poor. It is now an $81.4 billion spread trade. Hospitals buy drugs at 20-50% below market. Bill insurers at full price. Keep the difference. The program grew 23% in one year. 15x since 2010. Independent oncologists cannot participate. They watch their patients get absorbed by systems that profit from a discount they are legally barred from accessing. 340B is a government program for the poor that made hospital executives rich.