mdp2021

27 ms·

421.
by mdp2021 2mo ago
The second "what could possibly go wrong" article I meet in this session. The other one is a piece where MIT researchers have modified cells to perform the function of transistors, and can structure a biological computer: «If deve
422.
AMD Working on a New Back End for Improving ROCm Compute in QEMU/VMs(phoronix.com)
2 points by mdp2021 2mo ago|0 comments
423.
by mdp2021 2mo ago
Very interesting and very good (though, let us say, much more recent that ancient): only, it paradoxically does suggest that the computational cost (when compared either to a simple "search by vectors distance" or to the task fail
424.
by mdp2021 2mo ago
> that's what you're [...] his definition that had lost No, again the point is not clear. When people mention a concept, they are assumed to know it. If the concept is "third world", that corresponds to the page o
425.
by mdp2021 2mo ago
We are having a communication failure here. You wrote, see the emphasis, «[/]superior[/] first world» in apparent reply to member flohofwoe's «former Soviet scientists and engineers trying their luck with 3rd world despots».
426.
Flexible brain circuits can switch between different tasks(news.mit.edu)
2 points by mdp2021 2mo ago|0 comments
427.
by mdp2021 2mo ago
There is a possibility that some just did not care about the whole financing endavour. If you don't particularly care for the life of cows, "looked a bit sad" and other false positive triggers become a sufficient marker for s
428.
by mdp2021 2mo ago
Which will have side effects affecting other realms (search and indexing etc.). And besides, when a regime asks e.g to "prosecute all intellectuals - wearing glasses will be a sufficient criterion to identify them" workarounds are
429.
by mdp2021 2mo ago
The point clearly is that they got reasons to avoid the term 'female'. The point is in the twisted absurdity brought by the evil.
430.
by mdp2021 2mo ago
No, it's informative. And a mention of morons being a transversal party of just occasional color, and a phenomenon that we have not finished assessing - a root cause in the shape revealed in this topics - is important here.
431.
by mdp2021 2mo ago
Thank you, probably so, but in that case I cannot see how said comment is consistent in the branch ("about file formats") - I do not get how it would follow.
432.
by mdp2021 2mo ago
(And/)Or maybe uncaring ("psychotic", technically speaking) exploiters of the system of incentives... "Conformists".
433.
by mdp2021 2mo ago
Yes, but no difficulty to believe it (in an informal Bayesian framework). Also reported interview experiences around the world in the past many years point to the same mentality - questions like "How do you believe you can increase [fe
434.
by mdp2021 2mo ago
> previous editions... used a sliver of the resources Well of course, but given that the previously employed technique was blitting, fliedumping memory areas, you can't be more efficient than that. Using XML is for transparency (r
435.
by mdp2021 2mo ago
Don't we have benchmarks for thinking quality - assessment over the "reasoning" output (correctness, structure, efficiency...)? We definitely should. And before the benchmark of the finished LLM, it would be interesting to co
436.
by mdp2021 2mo ago
Have you tried with different amounts for the "reasoning_effort (xhigh|medium|low)" parameter? Or the "<|think_xhigh|> | <|think_low|> | <|think_off|>" tags: apart from this template detail, it is not
437.
by mdp2021 2mo ago
> a third explanation for the term There is only one, only complex, that can be rephrased at will - and which was the intention of the coiner of the expression. > I was unaware that he coined that term in French (when it may alread
438.
by mdp2021 2mo ago
If we want to implement Intelligence, and especially now that "the box is open" we must, we can play with the "intuitive" LLM architecture to understand it and squeeze it to its potential yeld, but at some stage we have
439.
by mdp2021 2mo ago
> it's all human decisions. especially at institutional level, there's no causing others to act And what is the difference? It seems the same ("A got dangerous, causing B to act" vs "A got dangerous, having B d
440.
by mdp2021 2mo ago
What is the value in reminding the readers? Is there anybody who does not know at that level? Avoidable also in view of the following: have you noticed in the past years a surge of indoctrinated that, at every misdeed from some part P immed
441.
by mdp2021 2mo ago
> rather than political It is certainly political: "the countries non aligned in the "cold war", currently hindered, will demand a political weight". Why have you quoted me omitting «that will demand a better positi
442.
by mdp2021 2mo ago
> know that they don’t know And we are waiting for architectures that do - because it's duly.
443.
by mdp2021 2mo ago
> As to your definition of what "third world" meant, that's not how I remember it And the reading you adopted is a simplified superstition. The expression 'Tiers Monde' was coined by academician Alfred Sauvy in
444.
by mdp2021 2mo ago
Or key industrialists.
445.
by mdp2021 2mo ago
And how can your point be credible with the sweeping generalization and ignoring for the sake of the monothesis e.g. the systemic failures that brought us to this point. I'd start again using quantifiers and a more complexive analysis.
446.
by mdp2021 2mo ago
> I don't believe that's CiM as you advocate You missed the whole point of Taalas HC1: that it is Compute-in-Memory. > 2. Merging storage and computation // Modern inference hardware is constrained by an artific
447.
by mdp2021 2mo ago
Ok, high-speed ROM could be on the horizon. But ROM has a massive disadvantage being static. So, either it is cheap and practical "like a CD", or decision making will be forced to do its evaluations. We have a von-Neumann architec
448.
by mdp2021 2mo ago
> but this doesn't need the tool to be reliable as you said, just the solution part And where are the warranties that the solution built be reliable (and optimal, etc.)?.
449.
by mdp2021 2mo ago
> AI is supposed have solved the Which AI? LLMs are coding facilitators and code producers. A problem is solved when the solution is reliable. Non-deterministic Neural Networks are not reliable. In fact, > more mechanical[] task t
450.
by mdp2021 2mo ago
> Is llama.cpp (and thus llama.app) llama.app is just an URL (for the "advertisement" webpages of llama.cpp outside GitHub). > that much better than Ollama llama.cpp is the real thing, ollama was a fork that remained inf

More ›