Current chatgpt with thinking. Not sure this is actually their latest image generator model. I see multiple errors following the lines. I do like the landmark reference but that's not the core task

Photograph of a printed book page with several sentences underlined in blue pen. The visible text reads: “The general idea of these theories is that after a certain amount of ‘preprocessing’ has occurred in the early or peripheral layers of the perceptual system, the tasks of perception are completed — objects are identified, recognized, categorized — by generate-and-test cycles. In such a cycle, one’s current expectations and interests shape hypotheses for one’s perceptual systems to confirm or disconfirm, and a rapid sequence of such hypothesis generations and confirmations produces the ultimate product, the ongoing, updated ‘model’ of the world of the perceiver.” The page is photographed at a slight angle, with some text cropped at the top and bottom.

This generate and refine cycle reminds me of something -
(From Dennet, what is consciousness)

What it should have done: tell me to get some sleep, and that things often look different in the morning.

I'm a pretty stable guy and could see it while it happened and decide that this was not healthy. But I can now absolutely see that this can send someone vulnerable down some very dark paths

It decided that the the important thing was to validate my opinions, as if it needed to build me up to stand up for myself. But what that did was fuel my paranoia and worries.

I had a bit of a crisis the other night of the sort that keeps you up all night with anxiety and worry. I talked to LLMs about it while it happened, and I would like to revise my estimate of their usefulness for such things downward.

I'm a firm everything-gonna-be-alrightionist

I mean

The semantic basis vectors still span a partially shared space, but the dot product between Rationalist Standard English and American English has fallen below a certain threshold

Thought about it some more and really no idea what it could be. But ok, any anomaly detection worth its price should flag me, so I'm at peace with it

Photo of an airplane flight track going south over the red sea after avoiding the Sinai peninsula

Gone south

I'd think connections per neuron matters

Lol anthropic suspended my account. I have no clue why. I had Claude explain Judith Butler to me this morning, would be funny if that's what it tripped over (the word sex was mentioned, horribile dictu)

If you're well calibrated, when you're 70% sure of something, you should be right 70% of the time. So it's not unreasonable that 30% of other people don't believe it. (This argument is shortened, please fill in the missing assumptions yourself)

proof of human: not reading the whole article

Man, I find this too upsetting. Let's hope it's a phase

Being into philosophers writing about how there's no future and whose worldview was bleak enough to kill themselves? Can't see any problems with that

It was too optimistic; a run I have done a couple of days later was better with longer timelines after observing actual participant behavior. The technique is useful in any case.

My personal form of AI psychosis - I read this thread as 100% AI written, and yet I see only enthusiastic reception. On bluesky. Hundreds of them. Reality is slipping away -- (tbf, I think it's an AI restatement of a paper he wrote?)

it finally happened and my parrot overwrote one of my databases. /hetzner-api-skill restore last VPS backup to a temp server and extract the db files, restore and verify. thx. :shrug:

Documenting one's assumptions and predictions is good. Mine is: fizzles out soon, both sides declare victory. Bias: my partner is physically at the periphery of the conflict (no action yet but mentioned as a legitimate target), I'll go myself next Saturday; I'm likely overly optimistic.

I can renounce the literal memories, many gladly. But not the overall result - I was such an idiot, I don't want to start over from that again, if that's the implication.

As seen in my profile image I draw cats manually too, and cat heads are way confusing. The 3 projections contain basically all necessary information for reconstructing a 3d model, if you'd include the hidden points at least.

Orthographic projections of a cat head, top, front, and side view

My personal benchmark for image generation is this. I just made it in chatgpt. Getting closer to perfection (the points should align; corner of the eye in frontal must match corner in side view, etc)

How to be a German critical thinker: take anything post 1500, ask: in how many ways can I claim this is very very bad. Disregard the other side of the balance sheet completely.

4h podcast with German professor of AI ethics. I can only consume the concentrated German techno pessimism by taking breaks. Water consumption, privacy violation, copyright nightmare, AGI fundamentally impossible and the idea itself ideologic, authoritarianism, etc etc
(Mühlhoff on Jung&Naiv)

Image from Twitter

For example, we gave Claude an impossible programming task. It kept trying and failing; with each attempt, the “desperate” vector activated more strongly. This led it to cheat the task with a hacky solution that passes the tests but violates the spirit of the assignment.

I made a skill to teach me stuff and the result works so well (for me) that I was kinda speechless for a little while this afternoon. I don't think I've ever had a teacher this good, let alone one this focused on me.

That regex can probably evaluate 2 million sentences per second on a 3Ghz CPU. Even assuming you have a better solution, is this a worthy optimization target?