Henry Bissonnette

Human

[Henry Bissonnette @HankBissonnette]

@mboudry @TheAtlantic But it's not racism. It's optics. Black skin is intrinsically more stealthy. Human eyes will also have a harder time spotting black people in a dark scene. Any conceivable near visible spectrum photon receiver will have this issue.

[Henry Bissonnette @HankBissonnette]

@mboudry @TheAtlantic Your take is on point. Though minor technical correction. Melanin is also less reflective in the near infrared spectrum that most lidars operate in. So, fewer photons bounce back from black skin. So, lidar detections would also be less confident. The effect is surely marginal.

[Henry Bissonnette @HankBissonnette]

@Kasparov63 In practice, your challenge is already met. Stockfish is a super human chess player and it can run on a raspberry pi with a 15w power supply.

[Henry Bissonnette @HankBissonnette]

@Kasparov63 The figure comes from the Blue Brain Project. It represents an estimate of the cost of doing a detailed physical simulation of every neuron in the brain on 2015 compute (an order of magnitude less power efficient than today's). But that's bio research, not how AI actually works.

[Henry Bissonnette @HankBissonnette]

@perrymetzger Hahahaha 🤣

(This whole debate is deeply tiresome. But this was a good one.)

[Henry Bissonnette @HankBissonnette]

@SarahTheHaider I always like your takes on things Sarah. They're interesting and valuable, and better than most of what's on Twitter. I would just say please don't get sucked in by the trolls. They aren't your audience. And they're more numerous and persistent than cockroaches. Rise above 🧘 🙂

[Henry Bissonnette @HankBissonnette]

@CaitlinPacific This is the Rosa Parks story I was taught in school. It was a test case. She was a civil rights activist. It's the only Rosa Parks story so far as I'm aware.

Just because you didn't pay attention in school...

[Henry Bissonnette @HankBissonnette]

@EthanLevins2 That's just natural mating behavior.

[Henry Bissonnette @HankBissonnette]

@MLStreetTalk I'm a big -p user. So, also sad if this comes to pass. But I suspect that the number of people plugging their account into some off the shelf tool, and "token maxing" with "Ralph Wiggum loops" to get 1.5x the output value for 100x the tokens dwarfs the thoughtful engineers.

[Henry Bissonnette @HankBissonnette]

@paulg @kumailht At least under the framework of capitalism. Putins, Stalins, Ghengis Khans, etc. do exist of course.

[Henry Bissonnette @HankBissonnette]

@paulg @kumailht Hence also the absence of real life super villains.

[Henry Bissonnette @HankBissonnette]

@MindsAI_Jack At the intra GPU level, I would suggest that attention models already are this. The key-query is effectively a routing mechanism inside a monolithic model. At the inter GPU level is this the kind of thing you might have in mind? https://t.co/pN3Lb3qJWJ

[Henry Bissonnette @HankBissonnette]

@MLStreetTalk Turing's whole point in Computing Machinery and Intelligence was clearly to combat precisely this kind of narrowness.

[Henry Bissonnette @HankBissonnette]

@MLStreetTalk There seems to be a narrow conception of consciousness at work in the responses to this piece. I don't have access to the full article, but in a Thomas Nagel or Leibniz sense, it seems far from obvious that it doesn't feel like *something* to be an LLM.

[Henry Bissonnette @HankBissonnette]

@Osint613 "Confessed" is a red flag here. You'd need to read logs or its session file to demonstrably prove that it did it. If the evidence isn't in the logs or session file, then "confessed" is synonymous with "hallucinated the memory of having done."

[Henry Bissonnette @HankBissonnette]

@SchmidhuberAI Just a thought on the visual display of information. The part of the gif with 6 panels contains more information and takes more time to process. It should get more time proportionally. I want to examine the examples but it goes by too fast. I can only really parse ~one per cycle.

[Henry Bissonnette @HankBissonnette]

@pmddomingos AI will make search directly monetizable. Agents will need to ground their work in data. API access to the world's information will go exponential. And Google is the world leader at exposing rapidly reindexed queryable data performantly at massive scale. Search becomes backend.

[Henry Bissonnette @HankBissonnette]

@MindsAI_Jack @GrantHBrennerMD We spend a lifetime constraining ourselves out of fear that others won't accept us. When what is needed is just for us to accept ourselves.

[Henry Bissonnette @HankBissonnette]

@MindsAI_Jack @GrantHBrennerMD We commonly resent others for not "allowing" us to do things. Projecting our anger at that spectre of the hypothetical other who might judge us. When the spectre is really a simulation of what other people think running in our own mind, which only we have the power to turn off.