mdp2021

17 ms·

481.
by mdp2021 3mo ago
You are talking about two different things. Yes, to update the blueprint for new models two layers will be updated. That is the NN. To instead update the data on which to operate you could use a RAG to query. (As in "the Pathfinder 2.0
482.
by mdp2021 3mo ago
> who exactly is going to ask an LLM to look at hotels in Montreal for them Anybody who has a specific informal query ("SELECT ... FROM ... WHERE has_carpark AND ... ORDER BY score(has_jacuzzi , walk_distance(...) ...) DESC")
483.
by mdp2021 3mo ago
> decade after decade That can only work when there is physical capacity for improvement though. > underappreciated effects of the AI boom and associated money is that it has strongly reinvigorated R&D in hardware Yes, absolut
484.
by mdp2021 3mo ago
> how far we are away from having It depends on AMD now. What was planned after the 8b was a ~30b, which is already sufficient (or more, when running at ultra-high speed).
485.
by mdp2021 3mo ago
> narrow and specialized use cases Such as Decision Making. /s You just can't set a high enough threshold of intellectual effort for critical decisions.
486.
by mdp2021 3mo ago
> What does it take to go from here to a model on a pcie card or an m.2 card It is already that. > Will "intelligence" become much like a gpu As an option among the implementations. > Are we a couple years away They
487.
by mdp2021 3mo ago
> It costs something like $300,000 for the hardware to run a model of that size You did not compute that as the cost for a speculative card from Taalas, right?
488.
by mdp2021 3mo ago
Are you sure? Source? (does not seem to be https://taalas.com/the-path-to-ubiquitous-ai/ , for example)
489.
Reasons I [Gary Marcus] wouldn't count Google out(garymarcus.substack.com)
3 points by mdp2021 3mo ago|1 comments
490.
Anthony Kenny [Philosopher] (1931-2026)(dailynous.com)
1 points by mdp2021 3mo ago|0 comments
491.
by mdp2021 3mo ago
> Is there any LLM from exactly one year ago that would be worth running? Bad perspective: consider the correction: "when are thresholds of sought quality reached"? Hence: not "is there a 10yo from last year that could c
492.
by mdp2021 3mo ago
Not necessarily: it is relevant to Taalas only if it is a compute-in-memory architecture. The Jalapeño mentioned («Anthropic is not alone in walking this path») in the article is still a classical Von Neumann architecture. And Taalas'
493.
by mdp2021 3mo ago
Can that be done when the whole idea is to store a multiplier into a handful of transistors?
494.
by mdp2021 3mo ago
Well, 50TB ROM Taalas HC1 style would be apparently a 400000b transistor system through a chip sized 2.5 meters on the side... :)
495.
by mdp2021 3mo ago
> A small model would be [mobile size] A ~30mm side for the HC1 tech for an 8b model (still unclear the planned HC2)?
496.
by mdp2021 3mo ago
> Dixie Flatline In case some did not know: also the movie (actually TV series) is finally happening. # Neuromancer - Official Teaser ( https://news.ycombinator.com/item?id=49055037 )
497.
by mdp2021 3mo ago
Compute the cost of producing n of them devices, imagine a fair price based on that, and see if that local, blazing fast card* can be an asset that could be replaced periodically. *(It's local: private files managing firm oriented. It&
498.
by mdp2021 3mo ago
https://taalas.com/h-content/uploads/2026/02/graph.png
499.
by mdp2021 3mo ago
They were specializing their forthcoming system on 4-bit FP - which I understand is a structural decision. Bonsai Ternary (1.7bits/weight) is a compromise, compromise that has to make sense in the context - efficient when translated in
500.
by mdp2021 3mo ago
Well, given the data above, roughly a 220b transistors chip for the HC1 tech.
501.
by mdp2021 3mo ago
Not too dissimilar to the first HC1 (6nm 815mm² 53B Transistors embedding an 8b LLM): > Our second model, still based on Taalas’ first-generation silicon platform (HC1), will be a mid-sized reasoning LLM
502.
by mdp2021 3mo ago
The basis of Taalas is "compute in memory" electronics - past Von Neumann's separation of processor and memory. You need to be able to add|mul where the data (the weights) are stored.
503.
by mdp2021 3mo ago
> A great deal of shops in the Netherlands don't take cash There is a possibility that such practice violates EU legislation in the terms of "paper" Euro being legal tender on the territory.
504.
by mdp2021 3mo ago
> stopped using cash You stopped using cash? It is one of the most insane things I ever heard. You transactions are tracked, I assume? The rule in the eu (PSD2) is that a dozen private/public/hybrid/other databases, of u
505.
by mdp2021 3mo ago
> too few apps Temporary. The age of LLM aided development is current.
506.
by mdp2021 3mo ago
> What’s with fascination with agi/superintelligence We had relative strength, we needed more strength, we built cranes; there was something like intelligence, we needed more intelligence, we sought to fill the need; we were right
507.
by mdp2021 3mo ago
> what you mean by Only, simply, non technically, that Sanskrit literature happened, then (non-committal irrelevant sketchy for-purpose-of-representation "then") Greek literature happened and you can find the Sanskrit terms m
508.
by mdp2021 3mo ago
> but there is no "I" in repeated inference calls to large language models We do not assume that. We do not assume any "I" as any "persistent flow" with the current architectures. We find it bewildering th
509.
Headlights got brighter, whiter, and more blinding after dark(arstechnica.com)
9 points by mdp2021 3mo ago|0 comments
510.
The Youth AI Privacy Act's Privacy Paradox(eff.org)
3 points by mdp2021 3mo ago|0 comments

More ›