jacobr1

24 ms·

361.
by jacobr1 3y ago
We are still in a exploration area where the use cases are in flux. If you find a set of use cases that work for you, then it will be a boon, but if not it is worthless. These tools aren't magic black boxes that do everything. On the o
362.
by jacobr1 3y ago
This is a partial misunderstanding - the institutional purchasers very much are using debt. But at the level of making an offer, they do so in cash without any financing contingencies. The models under which they can make profitable purchas
363.
by jacobr1 3y ago
Interesting - any other manufacturers in that area with checking out?
364.
by jacobr1 3y ago
s/do/do not/ dry up
365.
by jacobr1 3y ago
Depends on how you define privacy. If definitely accessing the data to train it was illegal/wrong, then you can't. But if that is ok, then the question is to what application will the LLM be applied? A generalized chat interface p
366.
by jacobr1 3y ago
Not really. The ideal is to have companies that are delivering functionality while still respecting user's data rights. Apple's products are probably the closest to this ideal, given their incentives aren't about harvesting u
367.
by jacobr1 3y ago
I recommend brother laser printers to everyone. They are like Toyotas: beasts that just keep running without issues for years. The cartridges last for a long time and also do dry up (a problem I had with inkjets due to infrequent printing).
368.
by jacobr1 3y ago
This is a good point, but the distinction isn't computational/human, it is something like deducible from current knowledge vs requires physical interactions/experiments with the world. You don't necessarily need humans f
369.
by jacobr1 3y ago
> while arcade pinball, at least historically, was designed to eat your money. It is a nuanced tradeoff. You need to make it entertaining enough so that players will continue to play, but also difficult enough that they regularly lose an
370.
by jacobr1 3y ago
A few considerations: 1) you don't need to fully mitigate the sea-level rise for this to be of value. Even a few cm might make a difference on the margin for certain areas. 2) The projects themselves might be a value if we really could
371.
by jacobr1 3y ago
Open Source hasn't really taken over the world, if you look at end-solutions. As a software developer I might use an open source database, but as end-user I'm probably not going to use open-source accounting package - but I will u
372.
by jacobr1 3y ago
On the other-hand, companies that fall behind their competitors because they are spending time on adjacent activities rather than leveraging new capabilities on the market will also lose out. It isn't clear at this time that LLMs fall
373.
by jacobr1 3y ago
> which means going back to pre-Elon content moderation Not necessarily, it depends how things get implemented. There isn't any reason Twitter couldn't build content classification tools that allowed advertisers to select which
374.
by jacobr1 3y ago
This is common in the US too, usually they need to file with their state. Going private will often remove the need to file publicly with the SEC though (though even then, that depends on things like how many shareholders there are, not the
375.
by jacobr1 3y ago
> VP of Sales though They all want to be CROs (Chief Revenue Officer) now.
376.
by jacobr1 3y ago
Who knows, but the question is more of unit economics than company profitability. Falcon could be profitable on a per-launch basis, and all the profit gets shuttled back into R&D.
377.
by jacobr1 3y ago
One risk I've heard of second hand. If you work on your personal-interests or hobbies ... then that might cease to be fun and just become "work." I've not had that experience (I've always found the projects in my ca
378.
by jacobr1 3y ago
> As the infamous saying of Eliezer Yudkowsky goes: the AI does not hate you, nor does it love you, but you are made out of atoms which it can use for something else. My point is that this presumes the system reasons around the human res
379.
by jacobr1 3y ago
It isn't about learning prompt engineering ... but rather expanding your horizons and what applications are currently feasible with existing tools.
380.
by jacobr1 3y ago
> It's a web terminal that you type queries into and it returns results Even just chatGPT in its current form is more than that. It is a full-fledged virtual assistant - but an esoteric one. Using it as better search is one use case
381.
by jacobr1 3y ago
> It's not consciousness, free will, emotion, goals, instinct or any of the other facets of biological minds For most "doom" scenarios require only weaker assumption: AIs need to be goal seeking. If they can make decisions
382.
by jacobr1 3y ago
There is a longer-term problem of trusting the explainer system, but in the near-term that isn't really a concern. The bigger value here in the near-term is _explicability_ rather than alignment per-se. Potentially having good explicab
383.
by jacobr1 3y ago
That wouldn't pay off for paid accounts. The cost of paying for a netflix account, will be higher whatever share of royalties is allocated from that account to the rights-holder. If for no other reason than netflix themselves is taking
384.
by jacobr1 3y ago
I haven't yet read this paper, but other multimodal models have different layers for mapping the input/output of each modality to the common embedding space. So, to simplify, there is some high-dimensional coordinate that the audi
385.
by jacobr1 3y ago
The CS specific analogy might be, "Abstractions are leaky," or more broadly the, "map is not the territory." Any complex-phenomena that is modeled with with a simplification will have places where that simplification fai
386.
by jacobr1 3y ago
One reason I'm sympathetic to this view is how far behind pre-AI automation is. Whenever I work with small-businesses and governments (and bigco's too but in different ways) I am amazed by how much make work there actually is. Som
387.
by jacobr1 3y ago
It is grisly, but the role the government has here is in setting reasonable liability numbers for a life, currently somewhere near $750K on average in the USA. On top of that there are putative penalties (sometimes including criminal) for n
388.
by jacobr1 3y ago
Kindle Unlimited works[1] works that way. They have pool of royalties derived as a percent of revenue for the month, that then distributed by views on a proportional basis. They have a specific algorithm that defines a "normalized page
389.
by jacobr1 3y ago
Have you tried the new search that was just launched? It seems to have significantly improved search accuracy. I agree the old version wasn't that great (though still was one of the better options for finding usage of things in the wil
390.
by jacobr1 3y ago
Which is why Bing has the best head start. They have an MVP of combining up to date data from search and the latest in LLMs. I used it last night on a search that was roughly: "find the top 10 restaurants in mid-town Manhattan that wou

More ›