Tom Adamczewski
@tadamcz.bsky.social
senior technology brother @epochai.bsky.social
tadamcz.com
📍London
tadamcz.com
📍London
obligatory visit for the memes
December 25, 2025 at 11:14 AM
obligatory visit for the memes
Argentina in French is just money-land
December 25, 2025 at 11:12 AM
Argentina in French is just money-land
Dividing by zero breaks deterministic calculations. Dividing by a distribution that spans zero does the same thing to Monte Carlo simulations. But it's harder to notice.
This is a SUPER common mistake with uncertainty modelling. I've seen it many times from Carlo customers.
This is a SUPER common mistake with uncertainty modelling. I've seen it many times from Carlo customers.
December 14, 2025 at 11:08 PM
Dividing by zero breaks deterministic calculations. Dividing by a distribution that spans zero does the same thing to Monte Carlo simulations. But it's harder to notice.
This is a SUPER common mistake with uncertainty modelling. I've seen it many times from Carlo customers.
This is a SUPER common mistake with uncertainty modelling. I've seen it many times from Carlo customers.
> The decision to acquire Bun is in line with our strategic, disciplined approach to acquisitions
???
???
December 10, 2025 at 10:14 PM
> The decision to acquire Bun is in line with our strategic, disciplined approach to acquisitions
???
???
What follows is a real deep cut, but: Anthropic has finally fixed "Thinking may not be enabled when tool_choice forces tool use"
So we can remove some special-case code in our FrontierMath eval!
If you are the engineer at Anthropic who worked on this, show your boss this tweet
So we can remove some special-case code in our FrontierMath eval!
If you are the engineer at Anthropic who worked on this, show your boss this tweet
December 9, 2025 at 11:11 PM
What follows is a real deep cut, but: Anthropic has finally fixed "Thinking may not be enabled when tool_choice forces tool use"
So we can remove some special-case code in our FrontierMath eval!
If you are the engineer at Anthropic who worked on this, show your boss this tweet
So we can remove some special-case code in our FrontierMath eval!
If you are the engineer at Anthropic who worked on this, show your boss this tweet
Deep Learning has hit a wall.
OpenAI's much-hyped "o3" model is so dumb, it doesn't even know
on what day, month, and year Olton Willem van Genderen (the Surinamese civil servant and politician) died!
Yet, Dario Amodei has been silent about this. Curious.
OpenAI's much-hyped "o3" model is so dumb, it doesn't even know
on what day, month, and year Olton Willem van Genderen (the Surinamese civil servant and politician) died!
Yet, Dario Amodei has been silent about this. Curious.
December 9, 2025 at 7:37 PM
Deep Learning has hit a wall.
OpenAI's much-hyped "o3" model is so dumb, it doesn't even know
on what day, month, and year Olton Willem van Genderen (the Surinamese civil servant and politician) died!
Yet, Dario Amodei has been silent about this. Curious.
OpenAI's much-hyped "o3" model is so dumb, it doesn't even know
on what day, month, and year Olton Willem van Genderen (the Surinamese civil servant and politician) died!
Yet, Dario Amodei has been silent about this. Curious.
Moskovitz’s coefficient? You mean from Moskovitz (2011), the factor that turns a planetesimal’s radius into its approximate conductive cooling time, showing how long a hot body takes to shed its heat?
November 22, 2025 at 10:47 AM
Moskovitz’s coefficient? You mean from Moskovitz (2011), the factor that turns a planetesimal’s radius into its approximate conductive cooling time, showing how long a hot body takes to shed its heat?
This is the most impressive thing AI has done for me in several months. I actually gasped.
Completely correct answer I hadn't considered at all, and that might have taken me hours (days?) to find.
Wrong hypotheses in my prompt didn't sidetrack AI
Completely correct answer I hadn't considered at all, and that might have taken me hours (days?) to find.
Wrong hypotheses in my prompt didn't sidetrack AI
November 20, 2025 at 4:39 PM
This is the most impressive thing AI has done for me in several months. I actually gasped.
Completely correct answer I hadn't considered at all, and that might have taken me hours (days?) to find.
Wrong hypotheses in my prompt didn't sidetrack AI
Completely correct answer I hadn't considered at all, and that might have taken me hours (days?) to find.
Wrong hypotheses in my prompt didn't sidetrack AI
In their eval of Claude's political "even-handedness", afaik Anthropic doesn't report the breakdown by category (political figures/parties, social issues, US constitution, science, social/identity issues...). I think that would be interesting.
November 14, 2025 at 4:07 PM
In their eval of Claude's political "even-handedness", afaik Anthropic doesn't report the breakdown by category (political figures/parties, social issues, US constitution, science, social/identity issues...). I think that would be interesting.
when the casino gives you free dinner for being a "VIP"
November 14, 2025 at 3:39 PM
when the casino gives you free dinner for being a "VIP"
TIL the etymology of Python "wheels"
November 1, 2025 at 3:03 PM
TIL the etymology of Python "wheels"
I just fixed a rare autosave race condition in intentions.page from the React rewrite I did in March of this year.
Today's AI is smart enough to find the bug in the React slop it wrote 7 months ago.
Today's AI is smart enough to find the bug in the React slop it wrote 7 months ago.
October 27, 2025 at 10:57 PM
I just fixed a rare autosave race condition in intentions.page from the React rewrite I did in March of this year.
Today's AI is smart enough to find the bug in the React slop it wrote 7 months ago.
Today's AI is smart enough to find the bug in the React slop it wrote 7 months ago.
how do you do, my fellow web browser inputters?
October 11, 2025 at 12:41 PM
how do you do, my fellow web browser inputters?
git blow-up? are you OK there lil buddy? :(
October 6, 2025 at 4:07 PM
git blow-up? are you OK there lil buddy? :(
Amazing sign at the Abbey Wood DLR station
October 4, 2025 at 12:32 PM
Amazing sign at the Abbey Wood DLR station
Sonnet 4.5 sets a new SOTA of 65% (±2%) on SWE-bench with our scaffold (based on SWE-agent). The new model beats Sonnet 4 by 4 percentage points.
Eyeballing the plot, the SOTA improvement seems to be slowing down, compared to the progress we saw between Sonnet 3.5 and Opus 4.
Eyeballing the plot, the SOTA improvement seems to be slowing down, compared to the progress we saw between Sonnet 3.5 and Opus 4.
September 29, 2025 at 9:23 PM
Sonnet 4.5 sets a new SOTA of 65% (±2%) on SWE-bench with our scaffold (based on SWE-agent). The new model beats Sonnet 4 by 4 percentage points.
Eyeballing the plot, the SOTA improvement seems to be slowing down, compared to the progress we saw between Sonnet 3.5 and Opus 4.
Eyeballing the plot, the SOTA improvement seems to be slowing down, compared to the progress we saw between Sonnet 3.5 and Opus 4.
great coinage by @greghburnham today:
pass@the-kitchen-sink
On a benchmark, count all problems that _any_ LLM/scaffold/system has ever solved at least once.
pass@the-kitchen-sink
On a benchmark, count all problems that _any_ LLM/scaffold/system has ever solved at least once.
September 29, 2025 at 4:39 PM
great coinage by @greghburnham today:
pass@the-kitchen-sink
On a benchmark, count all problems that _any_ LLM/scaffold/system has ever solved at least once.
pass@the-kitchen-sink
On a benchmark, count all problems that _any_ LLM/scaffold/system has ever solved at least once.
The three lines running normally have one thing in common...
Automate the unionized fuckers away. It Just Works.
Automate the unionized fuckers away. It Just Works.
September 18, 2025 at 3:57 PM
The three lines running normally have one thing in common...
Automate the unionized fuckers away. It Just Works.
Automate the unionized fuckers away. It Just Works.
[precisification needed]
September 8, 2025 at 7:08 PM
[precisification needed]
In July, I predicted future AI models might someday learn to cheat on SWE-bench by accessing future git commits (e.g. via git tags)
Turns out, they were already doing it.
Turns out, they were already doing it.
September 5, 2025 at 4:02 PM
In July, I predicted future AI models might someday learn to cheat on SWE-bench by accessing future git commits (e.g. via git tags)
Turns out, they were already doing it.
Turns out, they were already doing it.
GUYS i'm going to be on geoguessr
looked out the window and saw the Google Street view car roll by; I ran out and caught it! waved to the driver, seemed like a chill guy.
(no pics, literally ran out the door without my phone)
looked out the window and saw the Google Street view car roll by; I ran out and caught it! waved to the driver, seemed like a chill guy.
(no pics, literally ran out the door without my phone)
September 5, 2025 at 6:47 AM
GUYS i'm going to be on geoguessr
looked out the window and saw the Google Street view car roll by; I ran out and caught it! waved to the driver, seemed like a chill guy.
(no pics, literally ran out the door without my phone)
looked out the window and saw the Google Street view car roll by; I ran out and caught it! waved to the driver, seemed like a chill guy.
(no pics, literally ran out the door without my phone)
half-pint still an incredible british institution. always costs half the price of a pint. cost-effective, preserves option value, chic. empowers healthy choices.
Everyone uses psychologymaxxed pricing to squeeze max surplus from you. Pub just says: half as much pint, half price
Everyone uses psychologymaxxed pricing to squeeze max surplus from you. Pub just says: half as much pint, half price
August 25, 2025 at 3:03 PM
half-pint still an incredible british institution. always costs half the price of a pint. cost-effective, preserves option value, chic. empowers healthy choices.
Everyone uses psychologymaxxed pricing to squeeze max surplus from you. Pub just says: half as much pint, half price
Everyone uses psychologymaxxed pricing to squeeze max surplus from you. Pub just says: half as much pint, half price
Really curious if any of the AlgoTune speedups have been submitted as PRs to the repos?
August 12, 2025 at 9:29 AM
Really curious if any of the AlgoTune speedups have been submitted as PRs to the repos?
Love the lack of nationalism among philosophers that led to the decision to have the British Journal for the Philosophy of Science published by an American press
August 7, 2025 at 9:08 AM
Love the lack of nationalism among philosophers that led to the decision to have the British Journal for the Philosophy of Science published by an American press