xitter.cf/DanIsBuilding/status/2… it's very poor from them. We pivoted stacks and plans based on acceptance into this.

Claude Startup Program Explain to me how Anthropic can award startups thousands of dollars worth of seats, vouchers etc... give them time to plan months of business activity on that... then revoke them completely? And frame this as a 'pause' instead of a rugpull? @AnthropicAI

Got accepted with a legitimate application. The offers have all been removed from my account. It isn't a pause on applications. It's revoking something they agreed to from people who've now made business decisions based on what they were awarded. It's worth thousands of dollars.

Wait so the plans you'd awarded to us are being removed? My account no longer has them despite being legitimately accepted days ago...

I'm by no means an open ai stan, but this is total BS if your comparison is 6.1 Sol and Opus 5.5

ChatGPT Pro is officially recommending I cancel ChatGPT $200 and switch to Claude Max as it now wins on model quality AND raw inference offering... #chatgpt #openai

99% of developers want as much inference as they can get with the best models If OpenAI's plan is to drop out of that race and just wrap inferior models at inferior rates with software/bots, which you can build and host yourself using AI in an afternoon, they're finished #OpenAI

There is 0% chance this was fired from the hip. Far from it.

I assume this means refunds are available? Can't think of many things you can pay for where it's legal for the provider to retrospectively half the amount you get

Yep let's get one of them non human types ...

Except people that have had access for a week are saying it's worse than fable 5, stops tasks early and is bad at workflows and coherence. But don't let that stop your hype tweet based on one chart from the company that made it.

How do you know when each expires?

Asked 5.6 Sol High to perform a security review. Came back 8 minutes later to find it had burned through 100% of my 5hr usage having achieved nothing. I understand it's a more expensive model, but not 100x more costly - especially with the token efficiency gains? Anyone else?

How impressed you are by frontier LLM's ability to act, plan, decide or reason is inversely correlated to how much you know about the domain in question.

Was making the exact same argument to someone earlier. The act of coding always extrapolates from the intent. Unless your spec is literally one to one to the detail level of the resulting code (which as you put means it IS the code, or may as well be), you're trusting the agent to make decisions. That is fine if you go into it knowingly, but you can't spec you're way to a full proof LLM implementation.

Is posting daily to other people doing the same really building distribution?

Great summary. There are some issues with automated harness design, however. In most use cases, harnesses are super foundational and the cost of bad outcomes propagate to all of your agentic work within that harness thereafter. Unless your use case is mappable to a really clear reward function and your benchmarks, tests etc can capture a wide, but highly relevant, range of tasks, it's very difficult to get the speed and iteration advantage that you talk about. The problem then collapses into the quality and variety of benches you can build - which is arguably a harder task for most domains. I'm building a harness focused on helping people find and execute strategies for their personal development and success, based on their own skills and situation. You can measure outcomes, so instead you need to apply human judgement. So automated harnesses can make sense, but the domain being super narrow and measurable is key imo

Building in public

Does this platform still do the human tweets, or does that cost extra?

This is like when I vibe code something poorly, get stuck in a web of cut corners and start again from scratch a week later. Minus the trillion dollars.