Viewing profile — mdp2021

> 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).

> 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.

> 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…

> 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?

Are you sure? Source? (does not seem to be https://taalas.com/the-path-to-ubiquitous-ai/ , for example)

> 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 …

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 s…

Can that be done when the whole idea is to store a multiplier into a handful of transistors?

Well, 50TB ROM Taalas HC1 style would be apparently a 400000b transistor system through a chip sized 2.5 meters on the side... :)

> A small model would be [mobile size] A ~30mm side for the HC1 tech for an 8b model (still unclear the planned HC2)?

> 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=4905503…

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. …

https://taalas.com/h-content/uploads/2026/02/graph.png

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 m…

Well, given the data above, roughly a 220b transistors chip for the HC1 tech.

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 …

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 stor…

> 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 …

> 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 pri…

> too few apps Temporary. The age of LLM aided development is current.

> 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 intel…

> what you mean by Only, simply, non technically, that Sanskrit literature happened, then (non-committal irrelevant sketchy for-purpose-of-representation "then") Greek literature h…

> 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. …