"The moment you ask where the power comes from and when it arrives, the entire narrative falls apart in your hands."
I'm past tired of hearing about AI, but this person's engineering experience provides a different take than I've seen, and one that seems important and worth sharing to me.
h/t @composergreg, who shared this with me this morning.
youtube.com/watch?v=sQGZXrzykp…
#ai
- YouTube
Bekijk je favoriete video's, luister naar de muziek die je leuk vindt, upload originele content en deel alles met vrienden, familie en anderen op YouTube.www.youtube.com
Alison Wilder
in reply to Alison Wilder • • •The takeaway: the big AI companies are making ludicrous claims that are totally falsifiable about what their products will do to the job market. The incentive structure is such that creating fear around their products will make them loads of money.
I've heard that general take before, and it has felt intuitively true, but it's awesome to see someone provide evidence that the claims about job replacement are impossible without a quantum leap in energy generation etc.
#ai
ranjit
in reply to Alison Wilder • • •Yet another Josh
in reply to ranjit • • •@ranjit
That is one of the ideas, is that the power problem is already solved with some random research. It was grabbed, scanned, and ingested.
And you're only 1 prompt away from that answer.
We've seen this with some Erdos problems already. The answer to solve was already made, and an LLM jigsawed the completed parts together.
Alison Wilder
in reply to Yet another Josh • • •@crankylinuxuser can you expand on what you mean by "one of the ideas"? Do you mean that this is a defense against this guy's argument, or that people who are AI believers say this kind of thing to dismiss engineering arguments, or ?
@ranjit
Yet another Josh
in reply to Alison Wilder • • •@ranjit
As much as the AI hating squad doesnt want to admit... but LLMs are good at associating information the dense matrix was trained on. Now, LLM companies and such basically grab everything they can. Its probably upwards of a few exabytes of data ingested.
Now, within that sea of information and references and links.... real humans have solved problems. The problems may be in their own domain with their own language. BUT a tool that can probabilisticly compare across all fields of knowledge can show orthogonal ways to advance.
Now, these LLMs are NOT making new knowledge. But they excel at connecting the bulk of humanity's writings. And many of the problems we have is of association and connection.
Your initial comment of "quantum leap in energy production" is interesting. The idea is that it's already solved, kinda. Probably on paper, and/or already ingested into LLMs. But nobody's put all the pieces together in that exact "best way". BUT that
... Show more...@ranjit
As much as the AI hating squad doesnt want to admit... but LLMs are good at associating information the dense matrix was trained on. Now, LLM companies and such basically grab everything they can. Its probably upwards of a few exabytes of data ingested.
Now, within that sea of information and references and links.... real humans have solved problems. The problems may be in their own domain with their own language. BUT a tool that can probabilisticly compare across all fields of knowledge can show orthogonal ways to advance.
Now, these LLMs are NOT making new knowledge. But they excel at connecting the bulk of humanity's writings. And many of the problems we have is of association and connection.
Your initial comment of "quantum leap in energy production" is interesting. The idea is that it's already solved, kinda. Probably on paper, and/or already ingested into LLMs. But nobody's put all the pieces together in that exact "best way". BUT that LLM can potentially do it.
As an aside, I think both the techbros and the ai haters are both grossly wrong. These things do have utility (unlike the haters' comments), but are wayyyyyy oversold as to their capacities and inherent uncorrectable errors (techbros).
Alison Wilder
in reply to Yet another Josh • • •@crankylinuxuser gotcha. I'd say that's pretty similar in general to the take that the guy in the video expresses re utility.
I certainly would not assume that the problem has already been solved in pieces and that an LLM will put it together. Sure, LLMs have done that a couple of times, and impressively. But there are a lot more problems that they have not solved, so I don't see the logic in assuming they would solve the specific problem that makes this business not really that viable.
Yet another Josh
in reply to Alison Wilder • • •YEEEP and you didnt say the name of the phenomenon. Prolly cause its still too new.
Its the hope and promise of Jackpotting or Slot Machine and getting that exact right prompt and getting the answer out. And of course, @pluralistic has an article this exact name!
doctorow.medium.com/https-plur…
Its similar to the Vegas eyes-glazed-over smoker/drinker pulling a 1 armed bandit. But it is paying for more tokens on the hope that the NEXT question is the right one. Or on local LLMs, its the waiting for the response to complete.
And that next tranche of tokens COULD BE IT!
LLMs are slot-machines
Cory Doctorow (Medium)Alison Wilder
in reply to Yet another Josh • • •Yet another Josh
in reply to Alison Wilder • • •@pluralistic
Whats funny is this isnt the first time Ive seen this phenomenon. I first saw it in 2007 when I turned pagan.
Its all about #tarot .
Some folks I met would do a reading on the dumbest of shit. "Should I do $mundane_action?" . They'd paralyze themselves with inaction until some external force gave them approval, and that was tarot.
Whether you believe or not is irrelevant. Not trying to proselytize. But this has plagued folks learning divination all the time, and newer ones are especially more weak to it.
Ive called it Divinatory Paralysis. I see the parallels with asking the talking slot machine in the same way.
marieclaire.com/health-fitness…