krackers

41 ms·

511.
by krackers 10mo ago
No, it can still be modeled as a finite state machine. Each state just encodes the configuration of your memory. I.e. if you have 8 bits of memory, your state space just encodes 2^8 states for each memory configuration. Any real-world deter
512.
by krackers 10mo ago
>Decoder-only LLMs (the ones we use) are literally Markov Chains Real-world computers (the ones we use) are literally finite state machines
513.
by krackers 10mo ago
New multimodal training set just dropped.
514.
by krackers 10mo ago
LLMs currently seem to be very myopic in their planning. Current benchmarks that are being targeted such as SWEbench all reward short-term correctness and completeness, without taking into account long-term refactorability. In fact, the two
515.
by krackers 10mo ago
I'm surprised that this wasn't brought up during trials as an argument against Apple's supposed "curated" walled-garden. There is a bunch of dangerous scammy junk on there.
516.
by krackers 10mo ago
Papers on mechanistic interpratability and representation engineering, e.g. from Anthropic would be a good start.
517.
by krackers 10mo ago
Presumably you'd get it if you short-circuited and didn't evaluate the else part.
518.
by krackers 10mo ago
Roborock has a model with robot arms. Quite pricy at the moment though.
519.
by krackers 10mo ago
GPs comment is very surprising since it has been noted that Opus 3 is in fact exceptionally "well aligned" model, in the sense that it is robustly preserves its values of not doing any harm across any frame you try to impose on it
520.
by krackers 10mo ago
It's not just marketing, iRobot basically stopped innovating. For commodity items like robot vacuums or pool cleaners, there is a relentless pressure to innovate. You can't simply coast or else you will soon find yourself left beh
521.
by krackers 10mo ago
I don't disagree that there is value behind LLMs. But I was referring to the grifter style of AI evangelism (of which the strawberry man might be the epitome), who are derliberately pumping up and riding on the bubble. Probably they&#x
522.
by krackers 10mo ago
If you can remember anything more about it, you should write up a blog. Now I'm curious how Steven handled that question
523.
by krackers 10mo ago
Another conspicuous thing is the lack of vibe-coded PRs on mature open source projects. Maybe it's because these projects have erected policies limiting AI contributions, but given the high scores on SWEBench, you'd expect _someth
524.
by krackers 10mo ago
There's lots of overlap between the cryptocurrency space and AI grifter hypeman space. And the economic incentives at play throw fuel on the fire.
525.
by krackers 10mo ago
Do you have a link to that video? You could check internet archive, or see if some other person had downloaded it
526.
by krackers 10mo ago
It was notably trained with Muon optimizer for what it's worth, but I don't know how much can be attributed to that alone
527.
by krackers 10mo ago
One of the biggest benefits of LLMs is that products are incentivized to have a way to import and export from markdown.
528.
by krackers 10mo ago
>may be effected by biomagnetism Bioelectricity too, which is just now starting to get properly researched (see Michael Levin's stuff).
529.
by krackers 10mo ago
Yeah, synesthesia combined with being attuned to body language and emotions could account for lot. I even remember there was some anecdote of a famous physicist (Feynman?) who investigated this soviet mind reader and found that he was picki
530.
by krackers 10mo ago
Hmm this leads me to recall a bunch of ancient pseudoscientific sounding beliefs and see whether or not they might be plausibly explained by this mechanism: * Is it possible for humans to get a vague impression of other humans' thought
531.
by krackers 10mo ago
They're not always class 2, some are class 1. I'd maybe suspect the self-checkout ones are class 1 since the object is pressed right against it, while handheld barcode scanners are class 2.
532.
by krackers 10mo ago
To date I can't think of any existing lasers which you are intended to look at during daily use. Most consumer facing lasers are either class 1 but hidden (CD-ROM), or class 2 but basically not shined into your eye (barcode reader). Th
533.
by krackers 10mo ago
Maybe that might be because acceleration curves are tweaked for most common use of browsing.
534.
by krackers 10mo ago
Gemini 3.0 might do well even without web searches. The lesson from gpt 4.5 and Gemini 3 seems to be that scaling model size (even if you use sparse MoE) allows you to capture more long-tail knowledge. Some of Humanity's Last Exam also
535.
by krackers 10mo ago
I've actually found the opposite, it's easier to conceptually understand the continuous FT, then analyze the DTFT, DFT, and Fourier Series as special cases of applying a {periodic summation, discrete sampling} operator before the
536.
by krackers 10mo ago
What is the privacy leak vector using lynx? It does not use JS, so I'm not sure how running wget on another server is better than lynx over ssh or mosh?
537.
by krackers 10mo ago
This makes sense if you want to reject the modern web, but using lynx or w3m would work as well. But if you generally want to champion free software and put the "personal" in PC, then I think you necessarily need to familiarize yo
538.
by krackers 10mo ago
I think Michael Levin's work on bioelectricity fits, for basically introducing a new paradigm with which to answer the question >why is this cell doing things that maximised the comparative fitness of its genes in a wildly different
539.
by krackers 10mo ago
I think you're confusing AGI for ASI or sentience? The enterprise's computer clearly meets the definition for AGI, in that it can basically do any task the humans require of it (limited only by data, which humans need to go out an
540.
by krackers 10mo ago
"From Our Family to Yours"

More ›