troglodytetrain

At best this is a small piece IMO, but this does look like a piece of it, well done. At least based on my own experience exposing sys prompts from niche llm models in my saas niche. Should be higher upvoted. For those asking for proof, unfortunately it simply isn’t possible short of anthropic employee disclosure. All I know is that this is exactly how LLMs tend to leak: provide info they didn’t realize was sys prompt related and thus leak what they aren’t supposed to leak within their own current context window.

So when do we investigate USA companies for use of Chinese made products :)

So, no offense but DeepSeek and other Chinese providers are the ones actually advancing the price-perf here.

Being less expensive than Anthropic isn't impressive.

DeepSeek is amazing, they are, from a cost/benefit literally an order of magnitude or more better than the 'SOTA' models, and yet no one really talks about them.

I'm using them for my micro-saas, and they have made my niche economically profitable where as SOTA models are only slightly better for massively increased expense. Its truly impressive.

Word of advice to anyone, not all your use of LLM tech needs to be code/dev work related.

We are entering 'Web 4.0 era' or whatever you want to call it. Massive transformations of nearly every single business will and are being developed as the cost of intelligence as a commodity is falling through the floor...

This is very exciting, my own niche micro-saas has already been able to make heavy use of DeepSeek-V4-Flash for my use case, I am looking forward to seeing how performance improves.

So, at this point any company that doesn’t want ai agents navigating on their webpages have to kill the accessibility.

And they do and will. Blind users are a tiny tiny part of users. But hurting usability for the disabled is bad PR. So of course they do it “accidentally”.

How can they expect to A. Get away with political fallout of making local llm use illegal? B. Somehow enforce this? C. Do this without killing all scientific research in the USA (local llm seems to have a broad definition and I don’t trust any politician to get even close to understanding what it is) —-> If they somehow do this, all it will do is guarantee China pulls very far ahead.

Government solve problem. Problem gone. No more problem ever.

Problem come back? Because problem complex problem based on a large number of interdependent variables as is common is real life ecosystems? No way.

-Brought to you by grug gpt.

So, just be careful with this, it very likely is switching to other less capable model hence the cost reduction. So looks like Fable but isn’t. So you are doing extra work when you could just switch the model back to opus 4.8 instead.

So it is funny satire.

But my wife is finishing up her PHd and according to her: all students are using ai and pretending they are not because the PI and other older leadership treat any use of ChatGPT or similar as plagiarism and not allowed at any stage. Which I think is simply stupid.

I showed her how to use ChatGPT since Claude was blocking on all her request (biologist) and it has helped her massively improve the wording and structure of her paper as English is not a first language for her.

so just a quick counterpoint ---> I played around with qwen 2.5 4B recently locally. An older model and a tiny model.

Tested this model with optimized harness and it was good enough that it was able to complete (sometimes) a full job application flow for given json user data.

All that to say, I think that there is a strong possibility that we will even just with llm tech, develop it enough that local efficient models are entirely sufficient on their own to be able to drive basic robot 'harness'.

But we almost certainly will not be seeing that as the product from the corporations because they desperately want and need that data about you, for the same reason many tv (lcd/etc) companies make more money from 'ads' then from selling the tvs.

yeah, my personal work experience has taught me that extra productivity is actual really just a problem for many companies. You do more work now I have to too. We don't have to work because we have special relationships with or as gatekeepers ensuring we receive regular income from government, other corps, etc.

Look at what happened to twitter. So many companies aren't 'productive', they are simply parked somewhere where they can charge a toll, and then they pretend legitimacy.

So, maybe a dumb question, but do we have any real reason to believe that productivity at any level is actually associated with the actual underlying profit and earnings of any of the companies that are heavily 'knowledge work' focused, like big tech? Sure, for a factory, or a farm it makes sense.

But for tech I get the impression that valuation has been fundamentally unmoored from actual production for decades already. Its less about 'productivity' and more about vibes.

Another great product from anthropic where I need to pay hundreds upon hundreds of dollars for the privilege of being the unpaid QA team for the multi trillion dollar corporation.

Sounds like something that could be pretty useful as a 'validation' subagent. Provide it the details/context related to a larger LLM's run or turn in a harness and have it act as a gatekeeper. At this size and speed it looks like it could be economical to have it run every turn or even every tool call and inform the main agent about the result and success/failure.