lblume

13 ms·

211.
by lblume 2y ago
How would email be able to express anything the chat interface cannot? Attachments, tools, and long-form content are all well-supported. Email allows for asynchronous communication but that is usually not seen as a benefit - when I talk to
212.
by lblume 2y ago
Yes, it does. But the point is not that helping people today is pointless, in fact it may be all the more helpful because you take an active part in helping people attain a more positive life in the future. Conditional on an AI actually bei
213.
by lblume 2y ago
Of course it has fundamental problems. Likely they are more prevalent then in other techniques, regarding LLM training or otherwise. But one also needs to acknowledge how immensely useful and powerful the resulting models have become, achie
214.
by lblume 2y ago
I think this really comes down to personal preference. I also use Firefox on Linux and always find the Windows-type scrollbar to be incredibly ugly and bulky, especially on other overflow: scroll elements.
215.
by lblume 2y ago
Free speech should be considered a non-negotiable human right. You can't just trade those off against each other. The distinction into "materially meaningful" rights in my opinion just shows how easily both types of rights ar
216.
by lblume 2y ago
Good article, although the wider applications should likely be considered more. One small mistake seems to be defining Maybe (like in Haskell) as Nothing | Just but then still using Some (like in Rust) to initiate it.
217.
by lblume 2y ago
That is why all vaguely successful revolutions have been based on a Leninist-type vanguard party based on an educated "elite" claiming to lead the masses. For better or worse, their implementation of Marxist ideas was fairly consi
218.
by lblume 2y ago
Disagree. Class consciousness is about realizing that societal classes correspond to their relation to the means of production, not the concrete wage being earned. The majority of developers is part of the proletariat, at least according to
219.
by lblume 2y ago
Agree. It seems that despite model improvements our brains can still distinguish many kinds of AI generated text from human writing.
220.
by lblume 2y ago
Small LLM weights are not really interesting though. I am currently training GPT-2 small sized models for a scientific project right, and their world models are just not good enough to generate any kind of real insight about the world it wa
221.
by lblume 2y ago
By this point, it should be fairly clear that unregulated markets lead to monopolies. If modern libertarianism still entails "the freer the markets, the freer the people" I would consider this to be empirically disproven[1]. [1] E
222.
by lblume 2y ago
This looks very fancy, but some linebreaks in the statements would have really helped readability.
223.
by lblume 2y ago
Not necessarily. Recent advancements suggest that just a few soundbites of a typical person's voice already allow modern AI to synthesize it fairly accurately.
224.
by lblume 2y ago
How to classify that fraction? Let's say D. Gukesh (current chess champion) plays and learns chess for 50% of his time awake. Would you consider Stockfish to be a superintelligence that accounts for 50% of basic task wrt D. Gukesh? Wha
225.
by lblume 2y ago
When talking about superintelligence it is generally clear that general intelligence is meant, as specialized superintelligence has been solved for many areas for decades now without modern AI.
226.
by lblume 2y ago
For me, the second page redirects to the first. (And I don't live in France.)
227.
by lblume 2y ago
Exactly, same as panic! in Rust. There is a reason Rust was reluctant to add std::panic::catch_unwind at first. The docs thus explicitly mention that (1) it is not a typical exception mechanism and (2) that it might not even catch panics if
228.
by lblume 2y ago
Floats are, by their very nature, imprecise, so "excluding either" would not work up to a point. I would not want my choice of element to depend on the result of 0.1 + 0.1 (e.g. when multiplying by 10).
229.
by lblume 2y ago
No, the generic Google ToS ( https://policies.google.com/terms ) now apply universally and grant this kind of access.
230.
by lblume 2y ago
Web standards are moving fairly fast, and modern browsers are getting more and more complex in the process. It would likely be extremely hard, if not outright impossible, to build a browser completely from scratch today because of the sheer
231.
by lblume 2y ago
Well, we absolutely should be. The entire point of the open source community is to be able to react to this kind of organizational failure and disalignment with original values. Where exactly do the problems lie? Resource allocation? Trust&
232.
by lblume 2y ago
I agree that would be what good software would do, but what incentives would Spotify have to do that? They would have to pay for hosting and are possibly legally responsible for the file content, which could itself be illegal. And most peop
233.
by lblume 2y ago
> Unix philosophy. It's one day, one line. The Unix philosophy is often recited as "do one thing, and do it well". This does one thing, but doesn't do it well at all.
234.
by lblume 2y ago
Of course. It's just my opinion that this task would be perceived by most to be fairly repetitive and unfulfilling, but if the author thinks otherwise, great for him.
235.
by lblume 2y ago
Ok, obviously it's _doable_, but is it worth it? Using LLMs for this purpose would have been significantly cheaper, easier and with the right configuration just as reliable. Once the setup works, you could extend the analysis to all ki
236.
by lblume 2y ago
How exactly was the data evaluated? I would assume that manually checking every speech would be too labor-intensive?
237.
by lblume 2y ago
Am I missing something, or do the results not even look that much better? Referring to the output quality, this just seems like a different prompting style and RLHF, not really an improved model at all.
238.
by lblume 2y ago
I usually like my float representations to have some use. But sure, this does seem like a very interesting idea.
239.
by lblume 2y ago
> Any value representable in n bits is representable in n+1 bits > [1 1 1 1 1 1 1] 2.004e+19728 Does that mean that the 8 bits version has numbers larger than this? Doesn't seem very useful for 10^100 is already infinity for all
240.
by lblume 2y ago
> Far more reliable, no risk of hallucination. Only as reliable as the information contained in the system. The whole point of an LLM is to agglomerate as much information as possible with little manual intervention which causes them to

More ›