> That forces people to look beyond the walled garden. Every time I get "you used your quota, come back in 3 hours, or 2 days" -> that is experimentation time with their competition, leading to changed service plans. When they said "claude -p" will be billed at API pricing even for plan users I moved my harness off claude. After I integrated codex, then it was never going to be a full claude project again. What busin…
The claude -p thing was doubly stupid because they quietly allowed it again a couple of weeks later, so they pissed off developers for nothing
Wait, it's allowed again? Completely missed that. Been avoiding to use it and trying to find workarounds, not great.
Non-LLM user here. Why? Apart from the ecological issues, I'm very uncomfortable giving my organisation's crown jewels to {random_internet__corp}. Look at the lengths they go to for training data - 10M for Spirit's call logs? Destroying millions of obscure books to scan them? They make meth-heads look scrupulous. Your prompts, especially if they contain your entire codebase, are _way_ more data-rich than an airline p…
That's when you do the Enterprise deal.
I have first hand knowledge of household name companies with billion euro IP who use OpenAI and Athropic products quite liberally. They have WAY more lawyers than I do and their sole job is to keep the IP safe. They wouldn't sign a deal with even a slightest whiff of the IP being used to train anything.
And if it happens, the penalty for breach of contract would have so many zeroes it'd be enough to buy a country.
Open weights are certainly something that is very eagerly being adopted in some companies. It's not even that expensive because even the 15€/month/employee bill for people who use AI very little can stack up. Beside that though, many companies already have most if not all of their data in some cloud (Microsoft would be a prime example). Giving them a few extra bugs to get a (potentially) very useful tool is not out o…
>Open weights are certainly something that is very eagerly being adopted in some companies. A large local MSP that has significant colo space just started a big marketing push towards hosted LLMs for their corporate customers. I think its about to kick off everywhere.
It's not just putting up colo space for local models.
It's more about the harness and API proxy you use. They need to be smart enough to know when a (self)hosted model is enough and when to forward to a SOTA model.
The SOTA model _can_ do everything, it's just expensive as fuck. But so is shoving a difficult task to a sub-par model that takes (relative) ages and comes back with the wrong result.
Wow the sentiment here is so negative. I'm on the $200 plan (work pays) and I also have the $20 OpenAI plan (I pay) and keep a balance on OpenRouter. There is nothing as good as Fable, not even close. I recently had it run a 18 hour autonomous rebuild of a project (moving from Spark to Pandas for performance/data size trade off issues). It orchestrated Opus sub-agents flawlessly for 18 hours. It even did a great job…
the sentiment is negative and justified. anthropic nanny states what you can do. in your instance, anthropic may decide , arbitrarily, to stop 'autonomous rebuilds / refactors and ports' because they could pose some alignment/rights/etc risk to whatever slop their philosophers dream up while they're out eating $200 avocado toasts. then you can't do the thing anymore. fable is good, absolutely. agree it roasts Sol whi…
So which frontier AI company still has your goodwill and hasn't lost it?
It drove me to setup Qwen 3.8 this weekend. I couldn't see the value in just giving them money for a higher tier plan instead. I've never run a local LLM model before. Certainly won't take as long to iterate on this.
Qwen 3.8 is excellent. With the right harness, it does about 95% of what Opus can do, in my case automation software development. Since 3.8 came out, I have significantly revised my expectations for a local model. Give it another year or two, and we'll be running fast and free local models for nearly everything that matters, and these costly subscriptions will be a thing of the past. I've always believed that AI shou…
What harness would you recommend? I’ve tried Pi but the model struggled to stay on track after the compaction.
I have only 48gb of ram, so can fit only 80k context max, so good compaction is must.
> What I don't see is vast areas of industry finding $10s to $100s of billions of value in LLMs. Translation between languages. That value dwarfs all programming value that can be had. Economically, culturally, scientifically, spiritually.
Between human languages? The problem will be Google is giving that away for free.
You can do that with single words or phrases.
Now do a full movie's subtitles with google translate, in an automated fashion. So that it understands the context and still translates it correctly.
Anthropic lost all good will with me. Everything from their policies to their rhetoric has an air of "we don't trust you." The way they treated users who wanted to use them for OpenClaw didn't sit well with me, and then the Fable nonsense was the last straw. And they're somehow shocked users aren't loyal to that. It isn't about cost.
As soon as I started using Fable I was like, okay, this is probably as good a model as I will need for software engineering going forward. I still feel that way. I don’t need a better model, I need a faster Fable. The thing I miss most about programming is flow, and the constant bouncing between terminal tabs sucks. I’d love to do one thing at a time, with Fable, quickly.
There is GPT 5.6 Sol on Cerebras if you want to try that experience for an ungodly sum of money (not getting into GPT 5.6 Sol vs Fable, but only one is available on Cerebras) for an 11x speedup.
OpenAI also have 5.6 fast mode for a 2.5x speedup for 2x cost and is available on standard plans.
Non-LLM user here. Why? Apart from the ecological issues, I'm very uncomfortable giving my organisation's crown jewels to {random_internet__corp}. Look at the lengths they go to for training data - 10M for Spirit's call logs? Destroying millions of obscure books to scan them? They make meth-heads look scrupulous. Your prompts, especially if they contain your entire codebase, are _way_ more data-rich than an airline p…
> So they _are_ going to train on them, no matter how many checkboxes you tick to stop them. It is a risk, but is it a big risk? If one of the big labs were to do this and get caught it would be suicidal due to the loss of confidence in them and the inevitable lawsuits that would follow for breach of contract. Given the AI labs are all desperately trying to paint themselves as Serious Businesses so that other Serious…
No, how would they get caught? Even if their LLM outputs verbatim copies of the code, they can simply claim some victim company's employees bypassed the victim company's restrictions and must have used the code as input with an LLM. Joke is on you for doing business with them. And when there is some little known secret fact in the output, they claim it's been hallucinating... magical black box thinking makes it safe. It's a laundry for any input. A bit like a tor network routing for big tech deniability. Things go in, and things come out, but you can't prove the relationship between input and output as a third party, who isn't running the LLM.