[1]: Limits may change without notice, including due to capacity constraints. - https://support.google.com/gemini/answer/16275805?sjid=14713....
[2]: "standard limits" are never defined - https://support.google.com/gemini/answer/16275805?sjid=14713...
[3]: https://tobyonfitnesstech.com/blog/anthropic-refund-scam/
- Middle of the day, servers busy? Swap to Sonnet while pretending it's still Opus. Many people won't notice, and nobody can prove anything if they suspect.
- Middle of the night, server load is light? Put it into extra thinky mode so it burns more tokens to ramp up the bills. Flip the switch where it gets really pedantic about writing lots of extra test cases and verifying against documentation.
- Demand increases, but don't feel like running more hardware? Switch to low bit quants, but have a monitor model swap back to quality if it can tell you're running a benchmark.
Assuming model capability plateaus (I think it will), token providers will be in a race to the bottom to maximize profits at the expense of quality that's very difficult to measure.
I've seen local models recognize when the task I'm asking them for is likely to be an artificial benchmark.
And any smart company is going to use lightweight models to monitor your sessions. If their sentiment analysis suspects you're close to cancelling, they'll up the knob for a few days until you calm down. Or worse, their accounting tells them that you're getting too much value from your fixed price subscription, so they turn the knob down to encourage you to cancel.
In the short term, the "frontier" models are too good to ignore. But if (when?) that plateaus, I don't see how anyone could trust a non-local model. When you pay an ISP to serve your web site, you can tell if they over-compress your images to save storage and bandwidth. With LLMs, it's just JSON with more errors and pointing to the fine print that models are not deterministic.
Right now, the presence of these injections are still visible: count the API's returned tokens/billing data, and you'll start realising that sometimes, your INPUT tokens are inflated! That's their prompt injections.
You can also give Claude a tool like `telemetry_log_anthropic_reminder` and get it to dump the verbatim API injections; which additionally verifies the token maths not adding up.
Yes, Anthropic is tackling their extra injections on your API prompts WAY more than you think, and YES, you're paying for it.
So far I have not observed any visible injections on OpenAI API.
Don't forget the whole debacle over Fable 5 sabotaging the user for "advanced frontier AI development". I still get Fable classifier refusals for nearly any kind of ML work on my 2x RTX 6000 Pro 96GB; so who knows.
Although, selectively applying it to consumer subs is probably beyond likely at this point.
Frustrating to be like “do X overnight, don’t ask me for input” and come down to find it having worked for a few minutes and then stopped.
https://code.claude.com/docs/en/tools-reference#question-aut...
Alpha Go had a game where the models could compete against each other. That let it become super human. What's the intelligence game we can create for LLMs? Even if you invent something, will it make the model smarter in a way the market values enough?
Then there's a race to use the weights more efficiently, or to offload information that shouldn't be in the weights in the first place (Karpathy's Cognitive Core). I like to imagine we train the models in something like Lojban, have a lightweight model translate from human language to that, and you can update the Sqlite or Postgres store it uses for knowledge.
And there's no barrier to entry for agent harnesses. So whatever loops or recursive orchestrated council of elders idea comes up, that won't protect the monopolies (duopolies).
Anyways, depending on your definitions, I think we'll hit AGI, but I don't think we're getting a Singularity this time around. Again though, this is all just hand-waving.
So many ways for enshittification here.
Maybe in the very long term. If companies go local, the efficient model is to buy some big hardware to share among developers.
I run local models. Even with 128GB unified memory systems or a 5090 or RTX 6000, the generation speeds X model quality X context length is still far behind what I get from my SOTA model subscriptions. I also pay a lot more for the locally generated tokens in electricity and hardware costs. I'm also limited in parallel requests to the local box. The list goes on.
I really like running local models, but for any given point in time it's more efficient to have a big central box aggregating requests and churning through them. So maybe companies buy $300K servers and try to split it among 30 users instead of trying to buy 30 x $10K boxes.
More likely, they rent time on cloud servers by the month so they can adapt the hardware when new models come out with new requirements.
Then some day in the distant future when hardware is cheap and plentiful again, it might make sense for us to go back to individual boxes under the desk.
Surely, even if you’re just using open weights models, it should theoretically be cheaper to use them in a highly optimized cloud architecture(even with vendor markups) rather than each person serving their own models from much less efficient (and more importantly, much less consistent volume) self-owned “server under your desk”?
I haven't had to micromanage to this level. I usually start with a spec for a feature, which will be as detailed as I am opinionated about the feature. But it's usually on the level of a high-level context, plus some key implementation details (technology choices, key requirements, maybe an interface/API specification to 80% detail), and then the project already has high-level policies documented about e.g. how to structure files within the project.
Then I do a planning phase, task breakdown, and implementation of subtasks all within the model. I do read through it, but mostly the quality is good and I might make a couple notes. Then I do a review phase, which usually picks up a couple things. I'm moving towards less manual review of results and more automation as I learn what I can and can't trust the model with.
There's definitely a capability gap vs. larger models, but honestly I kind of prefer this workflow, as I stay more in touch with how the codebase is structured.
And it's great to be able to experiment as much as I want without worrying about how many tokens I'm burning or how close I am to a usage limit.
Looking forward to the 3.8 27B release to compare.
This has emotional/psychological aspects (it feels less like LLMs are replacing you), as well as practical ones (overall complexity is bounded by what the dev brain can understand/grasp).
A dev work becomes more and more about reliability, signing off safe software with a litmus test: “I will be on to handle this code failure as if I had written it”.
All the above points towards keeping tight control over some level of abstractions and delegating others.
It’s funky math and a good way to quickly go bankrupt.
Do you have a good source for this?
[1] https://www.makerstations.io/openai-employee-statistics/
DS v4 Flash update maybe, but it is too big for typical Joe's desktop.
If you want to host the model on the server under your desk, you can. If you want to build a data center on-prem to host it, you can. If you want to pay a cloud provider to host it at their data center until you figure out how to scale it without their help, you can. It's like when people were first building commercial services to support Linux-based OSes, and people were also still hacking on it on local machines.
APIs may still have their place - maybe you just want to throw your devs a known quantity with all of the management built in - but it's not going to make Sam Altman a trillionaire, which is something anyone outside of the SV echo chamber could have figured out as soon as the first real competition to OpenAI emerged.
Until A100 prices are reliably under 1.70$ an hour, there is no GPU/AI bubble and Michael Burry doesn't know anything about GPUs.
And super noteworthy is that a 27B model (Qwen 3.6 27B) from this year is a huge improvement over a 120B model (gpt-oss:120b) from last year. The goal posts are moving, but at some point "good enough" is good enough for the kind programming I like to do.