mrbungie

22 ms·

331.
by mrbungie 2y ago
I mean, sure, you can model a simple linear regression fitted via Least Squares (pretty much what they did 200 years ago) with a one hidden layer feed-fwd Neural Network, but the theorical framework for NNs is quite different.
332.
by mrbungie 2y ago
The point here is that "democratically passed legislature" is not at people's service.
333.
by mrbungie 2y ago
Good-faith rationale: None. Rationale: More growth for less effort, even if there is no soul. Fitting for our times.
334.
by mrbungie 2y ago
Cool project!, alternative name for a fork: PlatanOS
335.
by mrbungie 2y ago
Google doing the "look how we can do this but you can't and you won't with our help" with more force than ever.
336.
by mrbungie 2y ago
Putting people in PIP or firing them, are not good experiences, period. Only socio/psychopaths would feel powerful and good about having peoples future in their hands, and not all managers are socio/psycopaths.
337.
by mrbungie 2y ago
Are those the same market pressures that made Google discard or repurpose a lot of working old search tech for new shiny ML-based search tech? The same tech that makes you add "+reddit" in every search so you can evade the adversa
338.
by mrbungie 2y ago
Yes, that's because a coin toss is not intrinsically random, but just pseudo-random due to its chaotic behaviour which is especially notable at relatively "extreme" starting conditions. But if the tosser were to control, mani
339.
by mrbungie 2y ago
I mean, no product is better than a shitty product in this case.
340.
by mrbungie 2y ago
Never seen the "operational overhead elimination" really happen in the wild. Sure, you lose the N Sysadmins, you gain at least N+1 SREs/Cloud/DevOps Engineers.
341.
by mrbungie 2y ago
LLMs are excellent for scripting: be it python, shell or SQL, and you need a lotta scripting at any kind of job related to data, even when said scripts are just an enabler for delivering the pursued value. Total game changers in that space.
342.
by mrbungie 2y ago
What I would expect is a lot of "non-idiomatic" Go code from LLMs (but eventually functional code iff the LLM is driven by a competent developer), as it appears scripting languages like Python, SQL, Shell, etc are their forte. My
343.
by mrbungie 2y ago
I read "python 2" and a part of my soul cried in utf-8.
344.
by mrbungie 2y ago
A typical CEO's job is to guard and enforce a narrative. Great ones also work at adding to the narrative. But it is always about the narrative.
345.
by mrbungie 2y ago
Not really about Theory of Mind, but in the same line, I remember the other day someone argued with me that LLMs model the world, rather than just modelling language (that may represent the world). I kept thinking about that problem and pla
346.
by mrbungie 2y ago
> You can get one of those base models from OpenAI playground under Competions as "davinci-002". It's as weird as you want. Yes, "davinci-002" was not rlhf'd, so it may be better for creativity and similar t
347.
by mrbungie 2y ago
Current propietary rlhf'd unmodified LLMs are bland and boring. I think that's because aligning them to (1) "be generally useful", (2) "be safe" and (3) "be creative" at the same time via RLHF is a di
348.
by mrbungie 2y ago
So for making an edit you're proposing to extract every (possibly partial) objects/subjects out of a picture, create 3d models out of them and then animate? And if I don't know how, first learn how to do it? A sleep-deprived
349.
by mrbungie 2y ago
Integrations are not OAI's moat as those are primarily UX developments that are kept by Apple/MSFT. Right now, and if they want, they can change some lines of code to get up and running with another provider, like Anthropic or wha
350.
by mrbungie 2y ago
Training and inference costs: Paying for GPU time in Azure datacenters.
351.
by mrbungie 2y ago
Replacing call center roles with this is something I can see happening with the realtime api + voice output. Radiologists, I'm not sure what we need is just image model finetuning + LLMs to get there.
352.
by mrbungie 2y ago
People who can't guide themselves without relying entirely on the app. Normally, I just glance at what action I should take next (i.e. continue straight for 2km and then take a left turn at X) and then follow the road signs to do so. B
353.
by mrbungie 2y ago
By this point and by that I've read, cloud computing preferences are as subjective as liking any other thing. I've read here and in a lot of other places that AWS IAM is the worst part of AWS. But I don't know what to believe
354.
by mrbungie 2y ago
Yeah, I know. I didn't say it was incredibly better in every bench, but there are reasoning and planning benchmarks it does much better (not just slightly) than Sonnet 3.5 and the rest right now, and that still that makes their offerin
355.
by mrbungie 2y ago
Last time I checked, o1 did much better in PlanBench that models from other providers indicating better perf in that area [1][2]. Probably that lead will rapidly decay as other players implement similar approaches to "reasoning"&#
356.
by mrbungie 2y ago
Need evidence for that, afair this is a highly debated point right now, so no room for "obviously". PS: Plus, most reasoning/planning examples coming from LLM based systems rely in bandaids that work around said LLMs (rlhf&#x
357.
by mrbungie 2y ago
It isn't some "soul", but I think parent is making the same point as Yann Lecun usually makes: You can't have "true intelligence" (i.e. akin to human intelligence, whatever that is as we don't really know
358.
by mrbungie 2y ago
No one from the list (also add Anthropic to that list) is SOTA right now, but maybe Apple wants to keep its options open as the gap closes between OAI and the others.
359.
by mrbungie 2y ago
I guess parent is talking about using PCA for doing something akin to Factor Analysis (and making interpretations over the components and the loadings / contribution from each variable towards each component). I have at least seen that
360.
by mrbungie 2y ago
At least OAI and MSFT offer "copyright shields" for enterprise customers. Say what you want about the morals associated to said offerings, but it sounds very interesting for companies that may want to use a plagiarizing machine fo

More ›