iknowstuff
iknowstuff on Hacker News
Here let me point you to the assertion you made:
>> Its not happening, at least not with LLMs
> Correct.
> Llm reasoning is done exclusively in the way of symbolic manipulation.
No, LLMs are not a classical symbolic system applying explicit rules to symbols.
The article asserts general guarantees of program correctness have fundamental limits, but it explicitly says those limits apply to humans too. While that suggests neither humans nor LLMs can reliably solve every planning or verification problem, it says nothing about which will do better on the software problems people actually encounter.
It certainly does not present a fundamental problem with LLMs as they exist today.
How does this make the case for humans forever exceeding LLMs in planning and architecture? It doesn’t explain why a neural net couldn’t learn to do better than humans on this task at all. It certainly doesn’t point out any deficiency of attention, transformers, layers, encoders, decoders, or how the layers or their inputs/outputs are set up.
Talking about current LLMs while trying to assert future impossibility, and not mentioning an actual architectural problem, is giving a hint of not seeing past your own nose tbh
The estimate for six French EPR2s rose from €51.7 billion to €67.4 billion between 2022 and 2023. EDF’s later estimate is €72.8 billion, about 41% above the 2022 figure, before the reactors have been built.
I’d bet you the cost will be $140B when it’s all said and done, but this website won’t be around in 2044 (assuming no delays, good luck with that). You can bet regulations won’t stop changing over the span of decades either.
I’d rather have 70GW of solar and batteries deployed in a few years, and replace them with even cheaper ones in a number of decades.
What exactly is the architectural problem that has you convinced LLMs cannot learn how to strategize and architect well after enough RL?
Eh, this is why networks ossify and we can’t have nice things like a maintained connection when switching from home wifi to 5G. MPTCP is important, eg Siri uses it I believe?
If you’re on Ubuntu you can upgrade by pulling the -43 version from the -proposed repo. Crazy they still havent shipped it to -stable.
Nuclear costs $16B/GW. Solar $1B/GW nominal. Batteries $1B for 1GW/4GWh. Figure out how much solar overcapacity and batteries you can get, within a year or two, for the price of a nuclear reactor deployable 10-20 years from now.
Have we seen a single deployment?
Well actually given the free fall in prices of renewables and batteries, not really.
Nuclear costs $16B/GW. Solar $1B/GW nominal. Batteries $1B for 1GW/4GWh.
Figure out how much solar overcapacity and batteries you can get, within a year or two, for the price of a nuclear reactor deployable 10-20 years from now.
Beeper allows multiple accounts from various apps in one chat app FYI
...when preallocating space for a file. holy clickbait!
Rise of local models? what rise? I run qwen3.8 flash next on my 200GB/s 128gb Strix Halo and even that is barely acceptable, big models require so much more. Nobody's running a local model on their power sipping phone. It's just a nerd's wet dream.
Even the worst ev driven off coal nets out to less emissions in its lifetime than gas cars. this has been known and well studied for like a decade. Keep up!
Also their LFP batteries are extremely simple and recyclable forever.
Yea unfortunately eg Ubuntu’s bashrc has a script which will make it drop colors regardless of the presence of ghostty’s terminfo. Its `screen` will also fail, so e.g. it will refuse to upgrade to a new distro release.
Lots of jank due to this custom terminfo.
rootshell is libghostty and free, doesnt strike me as vc backed
Maybe the proposal you saw was breaking up into crates, which is Rust's compilation units and can be parallelized. They're not microservices, they're just libraries which still combine to produce a single binary.
There's a bug causing NSStatusItems in MenuBar<->WindowServer to enter a continuous redraw loop after a number of hours of uptime which pins CPU at like at least 10-20%. I filed it as FB24528844
1. WindowServer sends the app a stream of remote-context notifications (CGSDatagramReadStream::dispatchMainQueueDatagrams -> remote_context_notify).
2. AppKit responds by rebuilding the status-bar window's tracking areas and structural regions (_NSTrackingAreaAKManager _updateStructuralRegionCreateIfNeeded / displayCycleUpdateStructuralRegions, including NSButtonCell _hasRolloverAppearance for the menu bar rollover effect).
3. That dirties the window, so UpdateCycle/stepTransactionFlush commits a CA::Transaction to the WindowServer.
4. Another notification arrives and the loop repeats at display refresh rate.
do you have alerts set up for your personal site's traffic?
hackernews is sleeping on this model
grid forming inverters : )
Yes! I had it refuse on a similar claim of legality and when challenged to say what law forbids it it proceeded. Lil bitch.
Nah sadly not. It's very slow, fails to respond half the time (just decides to disappear, ot "we ran into a problem"), it's nowhere near the full duplex voice of GPT Live - Siri is still some dumb text model behind a TTS model, and does not work like modern agents which can take chains of actions for me. It's pathetic.
Yeah if you pay for his patreon you can read his discord where he’s mentioned it. Also I think on stage at a recent youtube convention in san mateo he mentioned he always looks for the hook.
Note also the increase in video length corresponding with the platform’s algorithm changes
You’re incorrect about Alec from technology connections. He is very aware of hooks for videos and retention is very much on his mind. His videos are well scripted, for the most part, given his niche. But the reason the videos are an hour long is BECAUSE of the algorithm rewarding long form second screen content.
It's literally required by the government's PERM process