bob1029

41 ms·

841.
by bob1029 9mo ago
I think user interface design is a very cursed task for an LLM. They are skilled at giving you high quality CSS and design patterns, but they're horrible at actual layout and composition. I can say "add a status strip to the form
842.
by bob1029 9mo ago
Not relevant for modern LLMs, but concatenative, stack-based languages are very good at genetic programming applications. Concatenation tends to yield more viable programs when things are being mutated over time.
843.
by bob1029 9mo ago
> Use radioactive decay It's a lot easier to use diodes (light emitting and otherwise).
844.
by bob1029 9mo ago
I'm a bit lost on this one. I can get a ridiculous amount of tokens in and out of something like gpt-5.2 via the API for $100. Is this primarily about gas town and friends?
845.
by bob1029 9mo ago
I've changed my mind about the short lived cert stuff after seeing what is enabled by IP address certificates with the HTTP-01 verification method. I don't even bother writing the cert to disk anymore. There is a background thread
846.
by bob1029 9mo ago
Why not use both DLLs? Prefer win32 wherever possible and use the lower level APIs only if absolutely necessary. Benchmark after you have figured this out. Performance is probably not a thing at this level of abstraction.
847.
by bob1029 9mo ago
The analogy that an LLM is simply an amplifier is apt for most general business. If you've already got a very effective team with clear vision/goals, this technology will almost certainly help to some degree. If you've got a
848.
by bob1029 9mo ago
Ah yes. Forcing the user to interact so you can make them listen to trash music. Excellent web design tactic.
849.
by bob1029 9mo ago
> Doing things at compile time / AOT is almost always better for perf https://devblogs.microsoft.com/dotnet/bing-on-dotnet-8-the-i...
850.
by bob1029 9mo ago
Not all hardware is digital but we can solve most of the hard parts in the digital domain. There's no reason to do everything analog just because it starts or ends that way. A lot of RF design has been reduced to arrays of dumb antenna
851.
by bob1029 9mo ago
This is why we don't use diffusion style models for diagrams or anything containing detailed typography. An LLM driving mermaid with text tokens will produce infinitely more accurate diagrams than something operating in raster space. A
852.
by bob1029 9mo ago
The brain relies on discrete, sparse events in space and time to handle computation. Most of the computation and learning that occurs is attributable to the relative timing of spiking events. A lot of information can be encoded in the del
853.
by bob1029 9mo ago
They can also train models using this silicon. They're advertising 24T parameter models on their site right now.
854.
by bob1029 9mo ago
> restricting eating to a short window - often about eight hours 8 hours is not a short window. I'm pretty sure Golden Corral will kick you out somewhere around hour #3.
855.
by bob1029 9mo ago
> The only heat that would be different would be radiated heat, which your food won't have a ton of Your food won't radiate much heat, but your oven certainly does. Aluminum + broiler is like photolithography for food.
856.
by bob1029 9mo ago
If you step back and look at this whole thing from a marketing and cash flow perspective, I think it makes a lot more sense. It is in OAI's best interests to create a perception that flinging agentic swarm crap at the wall may result i
857.
Evolution "Doesn't Need" Mutation. Here's Proof. - Blaise Agüera y Arcas [video](youtube.com)
1 points by bob1029 9mo ago|2 comments
858.
by bob1029 9mo ago
> 84 / 154 commits (54.5%) were lock/claim/stale-lock/release coordination. Parallelism over one code base is clearly not very useful. I don't understand why going as fast as possible is the goal. We should be tr
859.
by bob1029 9mo ago
You can do a lot of "native" windows development from modern C#/.NET via win32 interop. Newer C# features like ref returns, structs, spans, et. al., make the overhead undetectable in many cases. https://github.com&
860.
by bob1029 9mo ago
Did the author claim this? OpenAI and Cerebras have been working together at some level for nearly a decade.
861.
by bob1029 9mo ago
The #1 predictor of success here is being able to define what success looks like in an obnoxiously detailed manner. If you have a strong vision about the desired UI/UX and you constantly push for that outcome, it is very unlikely you w
862.
by bob1029 9mo ago
I was never too fanatical about the signal chain. The room and the loudspeakers were always more important to me. There was a time when things like SACD and high res audio were interesting, but it became clear this stuff has negligible impa
863.
by bob1029 9mo ago
I like how far we took this idea. https://www.youtube.com/watch?v=7uc2JgNQN0o
864.
by bob1029 9mo ago
Diet is king but exercise is not to be underestimated in its contribution. I have found major success with fasted cardio. That is to say, timing when the exercise occurs relative to your most recent meal seems to dramatically modulate the e
865.
by bob1029 9mo ago
I don't disagree. However, the TUI seems to have become the final destination for some major AI vendors. This doesn't look like a stepping stone to me.
866.
by bob1029 9mo ago
I genuinely don't understand the TUI obsession for LLM applications. Go watch copilot drive VS2026 if you've never seen it in action. There is no way you are going to be able to communicate this same amount of information via plai
867.
by bob1029 9mo ago
It's the border that hurts my visual cortex.
868.
by bob1029 9mo ago
> Defeatable interlocks are allowed on electrical equipment up to 600 volts. Fun fact, "high voltage" starts at 600v according to many regulatory bodies. 1kv is another popular threshold. The blast pressure from an arc can get
869.
by bob1029 9mo ago
https://www.fincen.gov/resources/statutes-and-regulations/us...
870.
by bob1029 9mo ago
Gas plants are only the most expensive in the simple cycle configuration. Combined cycle plants are competitive with other forms of baseload generation. The trouble is the response time. With day ahead forecasting, we can try to turn that p

More ›