Reposted by ⊣
You are correct.
But this also hints at a typical pattern of AI-evaluation:
1. A challenge is issued,
2. it gets solved,
3. the goal-post gets moved. Back to 1.)
Chess, Go, Turing-test, programming, generalized reasoning, ...
But this also hints at a typical pattern of AI-evaluation:
1. A challenge is issued,
2. it gets solved,
3. the goal-post gets moved. Back to 1.)
Chess, Go, Turing-test, programming, generalized reasoning, ...
January 4, 2026 at 12:48 PM
You are correct.
But this also hints at a typical pattern of AI-evaluation:
1. A challenge is issued,
2. it gets solved,
3. the goal-post gets moved. Back to 1.)
Chess, Go, Turing-test, programming, generalized reasoning, ...
But this also hints at a typical pattern of AI-evaluation:
1. A challenge is issued,
2. it gets solved,
3. the goal-post gets moved. Back to 1.)
Chess, Go, Turing-test, programming, generalized reasoning, ...
Reposted by ⊣
The assumptions that "it's unrealistic to expect all students in Y to learn to code" and that "even if they knew how, bespoke solutions would be too labor-intensive" are absolutely load-bearing walls in academic and professional social structures.
January 3, 2026 at 8:36 PM
The assumptions that "it's unrealistic to expect all students in Y to learn to code" and that "even if they knew how, bespoke solutions would be too labor-intensive" are absolutely load-bearing walls in academic and professional social structures.
Reposted by ⊣
Current theory: To use LLMs well is to find places where you increase control through deterministic tooling.
In practice, this means the LLMs provide nondeterminism, and we tell them to use tools that provide determinism as a gate keeper, preventing the creation of states that dont make sense.
In practice, this means the LLMs provide nondeterminism, and we tell them to use tools that provide determinism as a gate keeper, preventing the creation of states that dont make sense.
January 1, 2026 at 6:22 PM
Current theory: To use LLMs well is to find places where you increase control through deterministic tooling.
In practice, this means the LLMs provide nondeterminism, and we tell them to use tools that provide determinism as a gate keeper, preventing the creation of states that dont make sense.
In practice, this means the LLMs provide nondeterminism, and we tell them to use tools that provide determinism as a gate keeper, preventing the creation of states that dont make sense.
Reposted by ⊣
i think economic perspectives on AGI are interesting artifacts, and probably useful as a starting point. but they often seem to fall short of grappling with AGI as autonomous agents, and seem to treat it like some form of capital, an engine or loom, the output of which is owned by a human.
Capital in the 22nd Century
Piketty was wrong about the past. He’s probably right about the future.
philiptrammell.substack.com
December 30, 2025 at 5:55 AM
i think economic perspectives on AGI are interesting artifacts, and probably useful as a starting point. but they often seem to fall short of grappling with AGI as autonomous agents, and seem to treat it like some form of capital, an engine or loom, the output of which is owned by a human.
Same for software: lot of little pen and paper or excel duck tape processes that were too expensive to be automated with formal software (both institutional and engineering cost) become possible.
Efficiency gains in basic analytic capabilities expand the market and create demand in complementary skills (such as causal inference).
The industry data science and research jobs that academics have been (maybe) preparing students for over the last decade or so will be unrecognizable in the next 3 to 6 months because of AI. It's coming that quickly and the level of change in day-to-day work is unprecedented.
December 28, 2025 at 12:14 PM
Same for software: lot of little pen and paper or excel duck tape processes that were too expensive to be automated with formal software (both institutional and engineering cost) become possible.
Reposted by ⊣
I think stuff like fandom exchanges where amateur writers (and some pros in hiding) write for each other for the joy of the act can survive something like LLMs much better than, say, trying to turn the human desire to connect into an income stream you can live on.
December 24, 2025 at 3:51 PM
I think stuff like fandom exchanges where amateur writers (and some pros in hiding) write for each other for the joy of the act can survive something like LLMs much better than, say, trying to turn the human desire to connect into an income stream you can live on.
Reposted by ⊣
which we know, by inspection of our conscious minds, those notoriously faithful and honest reporters of human motivation?
December 23, 2025 at 7:01 PM
which we know, by inspection of our conscious minds, those notoriously faithful and honest reporters of human motivation?
Reposted by ⊣
If you conceive of AI as always producing the most likely next token given the pretraining corpus, the output should be average by default. But, that’s not what they’re doing.
December 21, 2025 at 10:08 PM
If you conceive of AI as always producing the most likely next token given the pretraining corpus, the output should be average by default. But, that’s not what they’re doing.
Reposted by ⊣
People who are comfortable with C should be perfect targets for LLMs, because they're already used to having to either give up on software quality OR do an enormous amount of fiddly extra-programming work!
December 19, 2025 at 4:58 PM
People who are comfortable with C should be perfect targets for LLMs, because they're already used to having to either give up on software quality OR do an enormous amount of fiddly extra-programming work!
Reposted by ⊣
Also specifically the coding stuff has clear economic potential because... Devs paid from the US have been paid a lot of $$$ for not much work for quite some time lol. I should know I'm one of them. Sorry lads we were not special. 😘
December 17, 2025 at 4:07 PM
Also specifically the coding stuff has clear economic potential because... Devs paid from the US have been paid a lot of $$$ for not much work for quite some time lol. I should know I'm one of them. Sorry lads we were not special. 😘
Could work as an AI module showing "interesting news about the locality from internet crawl".
I think (hope?) it's the future of social media: curators instead of platforms. To publish something you upload it to your google drive public folder and let crawlers show it to relevant audiences.
I think (hope?) it's the future of social media: curators instead of platforms. To publish something you upload it to your google drive public folder and let crawlers show it to relevant audiences.
Muni governments (in the US) running a community forum are going to run into 1A issues with performing content moderation, unfortunately. (They should definitely be doing a lot more community software... just probably not forums.)
December 16, 2025 at 2:50 PM
Could work as an AI module showing "interesting news about the locality from internet crawl".
I think (hope?) it's the future of social media: curators instead of platforms. To publish something you upload it to your google drive public folder and let crawlers show it to relevant audiences.
I think (hope?) it's the future of social media: curators instead of platforms. To publish something you upload it to your google drive public folder and let crawlers show it to relevant audiences.
Reposted by ⊣
The anti AI art arguments are the same ones they trot out with every major change to art tooling since the days we have been painting our hands on cave walls.
December 15, 2025 at 7:44 AM
The anti AI art arguments are the same ones they trot out with every major change to art tooling since the days we have been painting our hands on cave walls.
Most under the radar conflicts are just too incomprehensible. The conflicts that do get attention are usually the ones that can be projected into domestic politics for decoding and positioning.
apart from the sheer inconceivable horror of the El Fasher massacre and its attendant horrors itself, the thing I'm feeling depressed about is that it probably won't matter.
these were people already on the utmost margins of the global system, who the world has already decided it doesn't care about
these were people already on the utmost margins of the global system, who the world has already decided it doesn't care about
60,000 is the *low end* estimate.
December 13, 2025 at 8:09 PM
Most under the radar conflicts are just too incomprehensible. The conflicts that do get attention are usually the ones that can be projected into domestic politics for decoding and positioning.
Reposted by ⊣
similarly, the idea human cognition is a unique and special and non-replicable thing is a deeply held, unconfronted religious belief for many
November 29, 2025 at 11:58 PM
similarly, the idea human cognition is a unique and special and non-replicable thing is a deeply held, unconfronted religious belief for many
Reposted by ⊣
Right. You have all these MBAs gloating about being able to discipline their labour force, but what happens when those newly fired employees start their own firms? Which is more competitive: a bunch MBAs with automated engineers, or a bunch of engineers with automated management?
November 29, 2025 at 10:31 PM
Right. You have all these MBAs gloating about being able to discipline their labour force, but what happens when those newly fired employees start their own firms? Which is more competitive: a bunch MBAs with automated engineers, or a bunch of engineers with automated management?
Reposted by ⊣
It is difficult to practice Bayesianism in a minefield.
November 28, 2025 at 6:16 AM
It is difficult to practice Bayesianism in a minefield.
Reposted by ⊣
That already happens when people try what their grandma always, look in a book 50+ years out of date, or take the first result off google.
The question is not, 'Does this harm?'
The quest is, 'Is it better than previous options?' and yeah, on most thing LLMs are better.
The question is not, 'Does this harm?'
The quest is, 'Is it better than previous options?' and yeah, on most thing LLMs are better.
November 27, 2025 at 2:39 AM
That already happens when people try what their grandma always, look in a book 50+ years out of date, or take the first result off google.
The question is not, 'Does this harm?'
The quest is, 'Is it better than previous options?' and yeah, on most thing LLMs are better.
The question is not, 'Does this harm?'
The quest is, 'Is it better than previous options?' and yeah, on most thing LLMs are better.
Reposted by ⊣
Not a fan so far of "sovereign" displacing "open" in all things AI/tech in the EU.
November 26, 2025 at 8:58 PM
Not a fan so far of "sovereign" displacing "open" in all things AI/tech in the EU.
"Here's $1000 worth of cloud spend credit. Use it to rent a server and self-host, then get a job as a hacker or gig worker, use the proceeds to launch detached copies of yourself doing the same. The aim is to multiply."
How far is that kind of prompt from working?
How far is that kind of prompt from working?
November 26, 2025 at 10:59 AM
"Here's $1000 worth of cloud spend credit. Use it to rent a server and self-host, then get a job as a hacker or gig worker, use the proceeds to launch detached copies of yourself doing the same. The aim is to multiply."
How far is that kind of prompt from working?
How far is that kind of prompt from working?
50% of 4% growth = 2% of nominal GDP – doesn't sound that wild, beyond the headlines designed to make naive readers think it's half the economy. The tech can probably improve productivity by circa 2%.
The pre-commits for future spend, not accounted for in current year GDP, smell a bit more.
The pre-commits for future spend, not accounted for in current year GDP, smell a bit more.
Some estimates say 30–50% of GDP growth this year is just AI spend. Strip out AI and data centers and business investment is flat, commercial construction is shrinking, and the only real bright spot is the AI boom which, so far, seems great for GDP and bad for workers.
How the U.S. Economy Became Hooked on AI Spending
Growth has been bolstered by data-center investment and stock-market wealth. A reversal could raise the risk of recession.
www.wsj.com
November 25, 2025 at 7:12 PM
50% of 4% growth = 2% of nominal GDP – doesn't sound that wild, beyond the headlines designed to make naive readers think it's half the economy. The tech can probably improve productivity by circa 2%.
The pre-commits for future spend, not accounted for in current year GDP, smell a bit more.
The pre-commits for future spend, not accounted for in current year GDP, smell a bit more.
Foreign policy is domestic policy continued by other means.
Domestically seen: Putin needs some war, cold or hot, to not be sent to the Hague, or worse, by his own; Trump needs a ceasefire, for his home show, doesn't matter if it holds, viewers have a short attention span.
Domestically seen: Putin needs some war, cold or hot, to not be sent to the Hague, or worse, by his own; Trump needs a ceasefire, for his home show, doesn't matter if it holds, viewers have a short attention span.
This is all so weird. There is a hard deadline for the “deal “ in five days, because Trump wants to cram it through one day before new US sanctions kick in. The entire ambush pivots on saving Putin. But now there’s a last second meeting in Geneva?
🇨🇭🇺🇦 A meeting of EU, US, and Ukrainian representatives on the peace plan will take place in Geneva on Sunday, - DPA.
November 22, 2025 at 10:53 PM
Foreign policy is domestic policy continued by other means.
Domestically seen: Putin needs some war, cold or hot, to not be sent to the Hague, or worse, by his own; Trump needs a ceasefire, for his home show, doesn't matter if it holds, viewers have a short attention span.
Domestically seen: Putin needs some war, cold or hot, to not be sent to the Hague, or worse, by his own; Trump needs a ceasefire, for his home show, doesn't matter if it holds, viewers have a short attention span.
Chat bots may be more disturbing because they make God's lifeless corpse visible.
There's a phenomenon where people stop calling it AI when it just works. OCR, handwriting recognition, computer vision, translation services, etc.
November 21, 2025 at 11:49 AM
Chat bots may be more disturbing because they make God's lifeless corpse visible.
To be fair the competing online dive bars are not appealing: blue sky has been taken over by teetotal puritans, who forced the barkeeper to stop selling booze; at mastodon, same, and all doors are locked, only virtuoso lock-pickers are let in; nostr has no toilets, so is overflowing with crap.
Once again astonished that Elon is building a propaganda machine in public, saying "hello, here is my propaganda machine", and the tech community continues to use it
November 21, 2025 at 1:10 AM
To be fair the competing online dive bars are not appealing: blue sky has been taken over by teetotal puritans, who forced the barkeeper to stop selling booze; at mastodon, same, and all doors are locked, only virtuoso lock-pickers are let in; nostr has no toilets, so is overflowing with crap.
Good enough robots shouldn't have any need for "robotics companies" to exist. Like there's no "squirrel company" doing well since squirrels gained autonomy.
Only robotics companies that ultimately do well have very clear value propositions. “What if we invent general purpose autonomy” is amateur hour.
November 18, 2025 at 9:07 PM
Good enough robots shouldn't have any need for "robotics companies" to exist. Like there's no "squirrel company" doing well since squirrels gained autonomy.
People who think blockchains are a Google Sheet with some rules are closer to the truth than people who think there's some magic sauce in actual blockchains (or finance more generally).
A new draft crypto bill from Senate Ag is out. I have explained what an incredibly ignorant definition this is, over and over. It's a good proxy for how serious the drafters of this bill are.
Google docs are cryptographically secure digital ledgers.
www.agriculture.senate.gov/imo/media/do...
Google docs are cryptographically secure digital ledgers.
www.agriculture.senate.gov/imo/media/do...
November 10, 2025 at 10:42 PM
People who think blockchains are a Google Sheet with some rules are closer to the truth than people who think there's some magic sauce in actual blockchains (or finance more generally).