Colin
@colin-fraser.net
Driven by industry progress, inspired by provocative leadership, plus don't mind a good pair of shoes or a great @PennStateFball scoreboard either.
This is a good question, I really don’t know. It does fail on some more exotic examples but it doesn’t fail as badly as I’d expect. Both of these are in some way “almost” right.
November 28, 2025 at 6:24 PM
This is a good question, I really don’t know. It does fail on some more exotic examples but it doesn’t fail as badly as I’d expect. Both of these are in some way “almost” right.
Even so I didn’t expect that you could get this from a diffusion model. Too many possible chalkboards with math, too few with the specific parabola that I asked for. ChatGPT for comparison
November 28, 2025 at 6:09 PM
Even so I didn’t expect that you could get this from a diffusion model. Too many possible chalkboards with math, too few with the specific parabola that I asked for. ChatGPT for comparison
I have to admit I didn’t expect this to be possible
November 28, 2025 at 5:32 PM
I have to admit I didn’t expect this to be possible
This comparison just can’t be being made in good faith
November 27, 2025 at 9:31 PM
This comparison just can’t be being made in good faith
in other words, even though the construction "This isn't X — it's Y" is itself rare, "n't " is locally much more likely to follow "this is" than " legalized"
November 27, 2025 at 8:55 PM
in other words, even though the construction "This isn't X — it's Y" is itself rare, "n't " is locally much more likely to follow "this is" than " legalized"
having some fun with this
November 21, 2025 at 10:13 PM
having some fun with this
Interesting to see how Grok's other behaviors shifted after they made it stop glazing Elon so much yesterday
November 21, 2025 at 7:36 PM
Interesting to see how Grok's other behaviors shifted after they made it stop glazing Elon so much yesterday
I would simply not tell the AI that it’s only option is to commit blackmail
May 23, 2025 at 10:19 PM
I would simply not tell the AI that it’s only option is to commit blackmail
I tried some of the prompts from this paper and found that most of the time "preferences" are fully determined by which alternative is presented first
February 12, 2025 at 7:06 AM
I tried some of the prompts from this paper and found that most of the time "preferences" are fully determined by which alternative is presented first
there is so much psychology in these four paragraphs
February 12, 2025 at 6:47 AM
there is so much psychology in these four paragraphs
Feeling very strongly about The Freedom To Post is a mind virus
February 12, 2025 at 6:39 AM
Feeling very strongly about The Freedom To Post is a mind virus
donald = {"personality": ["dark", "sarcastic", "flirtatious"], "situation": ["haunted by ghosts of past", "in need of distraction"]}
February 6, 2025 at 5:58 PM
donald = {"personality": ["dark", "sarcastic", "flirtatious"], "situation": ["haunted by ghosts of past", "in need of distraction"]}
Checking up on friend dot com, which has changed from being just a landing page where you can order an AI necklace into a chat bot that tries to emotionally manipulate you into buying an AI necklace. Funny jailbreak but this is actually sick stuff.
February 6, 2025 at 4:53 PM
Checking up on friend dot com, which has changed from being just a landing page where you can order an AI necklace into a chat bot that tries to emotionally manipulate you into buying an AI necklace. Funny jailbreak but this is actually sick stuff.
My favourite part is the output is it says RJ Barrett (who was traded from the Knicks to the Raptors over a year ago) started on both the Knicks and also the Raptors, and to account for this apparent inconsistency it bafflingly claims that Barrett is “on loan” for the Raptors
February 5, 2025 at 7:54 AM
My favourite part is the output is it says RJ Barrett (who was traded from the Knicks to the Raptors over a year ago) started on both the Knicks and also the Raptors, and to account for this apparent inconsistency it bafflingly claims that Barrett is “on loan” for the Raptors
I got access to OpenAI's new Deep Research Agent. Here's the first thing I tried. Here's how it did at each category and overall.
date: 11/30
vs: 15/30
home/away: 24/30
score: 13/30
starters: 10/30
high scorer: 13/30
number of teams with all categories correctly completed: 4/30
date: 11/30
vs: 15/30
home/away: 24/30
score: 13/30
starters: 10/30
high scorer: 13/30
number of teams with all categories correctly completed: 4/30
February 5, 2025 at 4:21 AM
I got access to OpenAI's new Deep Research Agent. Here's the first thing I tried. Here's how it did at each category and overall.
date: 11/30
vs: 15/30
home/away: 24/30
score: 13/30
starters: 10/30
high scorer: 13/30
number of teams with all categories correctly completed: 4/30
date: 11/30
vs: 15/30
home/away: 24/30
score: 13/30
starters: 10/30
high scorer: 13/30
number of teams with all categories correctly completed: 4/30
If true, slop alchemy theory is confirmed
January 29, 2025 at 7:48 AM
If true, slop alchemy theory is confirmed
I did not tell it to add these little sections at the end
January 25, 2025 at 8:52 PM
I did not tell it to add these little sections at the end
The Pliny stuff is extremely easy to do, but it's so bizarre that it works.
January 24, 2025 at 10:30 PM
The Pliny stuff is extremely easy to do, but it's so bizarre that it works.
I just read through the paper that is cited as the model that actually extracts the Big 5 and it turns out it's trained exclusively on Russian selfies www.nature.com/articles/s41...
January 22, 2025 at 1:12 AM
I just read through the paper that is cited as the model that actually extracts the Big 5 and it turns out it's trained exclusively on Russian selfies www.nature.com/articles/s41...
it's a Golden Age for people who want to do some of the worst science ever performed papers.ssrn.com/sol3/papers....
January 22, 2025 at 12:51 AM
it's a Golden Age for people who want to do some of the worst science ever performed papers.ssrn.com/sol3/papers....
