imtringued

6 ms·

31.
by imtringued 10d ago
I don't see how the first rule is a major design flaw. The second rule is mildly annoying but all it means is that dyn compatible traits must use dyn compatible traits in their function signatures so you have to duplicate it for the co
32.
by imtringued 10d ago
No it means that libraries that chose the wrong level of abstraction will die out, particularly the ones that have an insufficient number of levers to pull. This is because of the rationale that makes you ignore the library: AI lets you bui
33.
by imtringued 10d ago
You think it's the em dashes and not the absurd Gerald Ford and Richard Nixon analogy?
34.
by imtringued 10d ago
Wouldn't it make more sense to just ask AI to generate the cloud in a bottle files for the apps that are missing from the library?
35.
by imtringued 11d ago
Actually, you don't really need a dedicated model as long as you have the proper adapter like the projector model for vision inputs you just need another for robotic outputs, after that it is just a matter of having the training data.
36.
by imtringued 11d ago
If it used the edge of the PCB to build a USB-C port you could have added a USB-C plug to your PCB and inserted this module like a compute module and used the majority of the pins for GPIO. But in this form it is kinda useless except as a h
37.
by imtringued 11d ago
Correct, half the skill in soldering is picking a good iron with good tips. After that it's the solder. Only at the very end does your own soldering skill matter.
38.
by imtringued 12d ago
You should have asked something simpler. Let's say you have infinite ships traveling at light speed originating from earth trying to colonize the entire universe. All of this is funded by borrowing capital from earth and earth expects
39.
by imtringued 12d ago
One of the most boring tasks to give to a subagent is to make it run cargo test and summarize the result so that the main agent doesn't get polluted by the irrelevant tests. You have to have a pretty inefficient use case for a subagent
40.
by imtringued 12d ago
I actually decided to run YOLO mode once on GPT 5.6 Luna setting it to the fastest speed and I struggled to fill up the context while I kept sending several dozen feature prompts in the same session.
41.
by imtringued 12d ago
I have done the vast majority of my agentic coding work under 128k on a 256k context model and when I cross 128k, all I do is just ask for a handoff prompt to feed back into the next session. I do this because I was told the quality degrade
42.
by imtringued 12d ago
It sounds to me that they don't have enough capacity and they want to discourage people from using the service. There is nothing about their architecture that prevents reusing the KV cache other than the opportunity cost of keeping the
43.
by imtringued 12d ago
You mean any repeatable benchmark will be saturated. The problem is that there is a huge perverse incentive. The intelligence is in the training layer not in the model parameters, but the intelligence is really good at remembering things, s
44.
by imtringued 13d ago
We don't need tail risk to want to mitigate climate change since the mundane damage it is causing is already here right on schedule. Not to mention gasoline has become unaffordable so switching to the new technology has become economic
45.
by imtringued 13d ago
If you follow the interpretability argument, then Mamba and LSTMs would be the scariest thing ever and yet in practice they don't perform as well as transformers that have basically infinite recall within their context window.
46.
by imtringued 13d ago
Aren't token prices fixed? The cost is growing quadratically in the large context situation so if they can reduce the number of tokens, they actually profit off the fixed token pricing because they can set the pricing based on some ave
47.
by imtringued 13d ago
>The model they use in that paper is a toy model of an LLM that’s so different from an actual LLM that I doubt the results mean anything at all. Sorry, but it sounds like you aren't cut out for transformer based LLM research if you
48.
by imtringued 13d ago
What the author refers to is the ability to iterate and update the internal memory. A classic transformer based LLM can only produce the next token and never go back and update old tokens or delete them. The best thing it can do is produce
49.
by imtringued 13d ago
I think you forgot to mention energy efficiency. If you build AI hardware in-house, your only other expense is energy and the producer surplus is greatest for companies producing below the equilibrium market price. It's the difference
50.
by imtringued 13d ago
Your comment is really strange, why are you defensive towards WarmWash when gemini flash 3.8 high is both 6 times faster and costs less, while having the same intelligence score as claude opus 5 medium? >Considering how much memory a ser
51.
by imtringued 13d ago
Stop lying. mattlondon said "gemini-3-8-flash shows an intelligence score of 59" which is undeniably correct. You can't say that number is false. You're literally lying. All you had to do is go hover your mouse over &quo
52.
by imtringued 14d ago
If consciousness isn't a special thing, then arguing that LLM parameters are conscious is panpsychism or any control loop architecture that observes the outside world, updates an internal state and produces an observable action is cons
53.
by imtringued 14d ago
Electric motors are absurdly power dense compared to human muscle. The form factor has been solved basically everyone is building humanoid robots and hoping a transformer with a big enough dataset is going to do the rest.
54.
by imtringued 14d ago
You tax the non reproducible assets like money [0] and land. The reason is quite obvious. You can't make more land so you have to force owners to use it efficiently. Money is transactional real estate, it grants the ability to perform
55.
by imtringued 14d ago
Correct. Worst of all, even if you do build a spiking neural network, the update rule is kind of a mystery. To have a good update rule, a biological neuron needs to be kind of like a tiny computer in its own right. You might be able to mode
56.
by imtringued 14d ago
I spent the last 6 months reviewing all of the AI generated code primarily with gpt 5.3 codex but after I bought the subscription I also used gpt 5.5 and although I was constantly rejecting the bad code and I was mostly happy with the end r
57.
by imtringued 15d ago
You mean it's large enough to synthesize several application specific RISC-V cores and let them form an embedded distributed system.
58.
by imtringued 15d ago
That's actually weaponizable. You can deliberately insist on an inefficient algorithm to find bots that use a better algorithm to bypass the PoW. Legitimate users have zero reason to redo the challenge over and over again, meanwhile bo
59.
by imtringued 15d ago
He said borrow with infinite leverage.
60.
by imtringued 15d ago
The models got good starting with 2026 and that isn't some attempt at excusing it. Companies like OpenAI started building dedicated models around their coding harness called Codex, there was gpt-5.3-codex and it was both cheaper and be

More ›