Testing the sync between bluesky and my website here

ja3k.com/shortform

Making it a thread and including an image for testing

It kind of worked!
ja3k.com/short/178687...

Attaching a speaker that plays Radiohead on loop to my Figure

Actually just OK Computer

Saw this at a local church. I think I used Google for the first time in elementary school and my kids may never know what Google is

Actually they already know to say "hey Google" to the Google home

I feel really sad every time I see a post about unhelpful grandparents. My kids are really lucky to have 4 grandparents who are great at taking care of them.

Sad I moved so far away from my own parents. But they still visit often and are a huge help when they do

One of my opinions that the best in any given field are usually actually underrated. Because so few can actually distinguish the quality of the great masters. And what they pioneer comes to feel so ordinary

Maybe another reason is that people have to signal their erudition by not naming the literal most famous person when talking about the greats. No one will think you're very clever for saying Einstein/Newton is the best physicist or Gauss/Euler is the best mathematician

I don't even try to evaluate whether the starship launches are successful

It's striking how partisan it is. I can basically predict people's takes in advance

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

I have an urge to write a CLI spaced repetition software but know it'd be a huge mistake

Alright I'll do it. Here's the README: I'll have Mentat write the whole thing. But not until Wednesday. I'm on vacation

Alright I (mentat) did it

Alright I'll do it. Here's the README: I'll have Mentat write the whole thing. But not until Wednesday. I'm on vacation

Alright I (mentat) did it

I wonder if this jump is also explained by insiders aware of anthropic 's computer use announcement today?

I still think it's pretty unlikely

Interesting to read his mindshare section in light of Google's "monopoly"

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

Very interesting article to read in 2025. I guess he was right the OS business was unstable for msft (I believe the majority of their revenue is cloud/office?). But he was wrong that apple couldn't survive long-term as a hardware company

Interesting to read his mindshare section in light of Google's "monopoly"

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

26 years later and apple fans haven't changed

From: (I'm clearing out tabs)

Oh this is the article where Neal says: "emacs outshines all other editing software in approximately the same way that the noonday sun does the stars. It is not just bigger and brighter; it simply makes everything else vanish."

Very interesting article to read in 2025. I guess he was right the OS business was unstable for msft (I believe the majority of their revenue is cloud/office?). But he was wrong that apple couldn't survive long-term as a hardware company

Interesting to read his mindshare section in light of Google's "monopoly"

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

From: (I'm clearing out tabs)

Oh this is the article where Neal says: "emacs outshines all other editing software in approximately the same way that the noonday sun does the stars. It is not just bigger and brighter; it simply makes everything else vanish."

Very interesting article to read in 2025. I guess he was right the OS business was unstable for msft (I believe the majority of their revenue is cloud/office?). But he was wrong that apple couldn't survive long-term as a hardware company

Interesting to read his mindshare section in light of Google's "monopoly"

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

Oh this is the article where Neal says: "emacs outshines all other editing software in approximately the same way that the noonday sun does the stars. It is not just bigger and brighter; it simply makes everything else vanish."

Very interesting article to read in 2025. I guess he was right the OS business was unstable for msft (I believe the majority of their revenue is cloud/office?). But he was wrong that apple couldn't survive long-term as a hardware company

Interesting to read his mindshare section in light of Google's "monopoly"

Didn't realize it was so long when I started. Probably should have just closed the tab

Tbh I liked it more than his fiction but I've only read 70% of snowcrash

Next tab I have to read is about sqlite triggers

Oh one more thing: it's interesting how different his experience of the OSes is from mine. I take it for granted every os has an easily accessible CLI. My first experience installing Linux was Ubuntu in 2011 and it was easy and not that different from windows really.

Turns out the flapper wasn't even the issue! But gpt -5's second suggestion that the float was too low worked. And people say vibe debugging doesn't work

If you even care

WTF it's still running. LLMs have hit a wall huh?

If you even care

WTF it's still running. LLMs have hit a wall huh?

Just had the most obvious realization that I can't believe no one told me 2 decades ago: the fact that the determinate measures the scaling of the unit cube *implies* it's multiplicative

I feel really stupid rn. For sure someone told me and I wasn't paying attention

My new thing is doodling

After I started this one I realized I've seen it a million times

I wonder why this is such a common complaint about Claude Code. I've literally never had this issue with Sonnet in Mentat. I also saw far less reward hacking with 3.7 (though still some) then I'd expect based on complaints

To be clear we didn't do anything to suppress this behavior. I wonder if the framing of "you're not just mutating the local file system, you're making a PR" makes it more focused?

Even the rinse aid uses AI and you think we're not in a bubble?

I can't even dry my dishes without RAG

Next we need the ability for Claude to message first "Just curious if the code ran? Want to share any screenshots?"

nice

@__hand_banana actually nevermind. I guess inflation is different

@__hand_banana I guess people hate inflation but they love their stocks/homes going up

We should say our ages like version numbers e.g. I'm jake-30 (new)

I'm sorry but jake-30 will be retired soon and won't be available via chat or api

Literally every day

I should have retweeted this the day after the gpt-5 release. It was extra topical that day. Though it is topical literally every day

Idk why the gpt 5 vibe is so negative. The is like the first OpenAI release since turbo I was excited about

Tbh I'd basically counted them out

@granawkins Remember a year ago when you said AI agents were a nothing burger?

@granawkins Do you have no ability to remember the past or imagine the future?

We are not post-eval in the sense that rigorous evaluation of models no longer matters But we are post-eval in the sense that no evals that people talked about a year ago matter anymore. And there are no good evals to directly compare frontier models

Even benchmarks that don't look saturated e.g. swe bench are so low quality that further improvements probably aren't even desirable

The labs are most incentivized to produce high quality benchmarks but they aren't incentivized to publish: 1. No one will take "we're the best at our own benchmark" seriously 2. They help other labs improve

One funny episode was when @bio_bootloader published loco diff @alexalbert__ retweeted and then quickly deleted his retweet. No way to know why and probably the best explanation is on reflection he thought the work wasn't of a quality he wanted to promote...

But I like an alternative explanation: the benchmark is extremely good and extensively used within Anthropic. That's why they're the best at it. And he doesn't want other labs to get the idea to train on it

Locodiff tests something every codegen agent has to do: understand the state of a file after editing it several times

Even benchmarks that don't look saturated e.g. swe bench are so low quality that further improvements probably aren't even desirable

The labs are most incentivized to produce high quality benchmarks but they aren't incentivized to publish: 1. No one will take "we're the best at our own benchmark" seriously 2. They help other labs improve

One funny episode was when @bio_bootloader published loco diff @alexalbert__ retweeted and then quickly deleted his retweet. No way to know why and probably the best explanation is on reflection he thought the work wasn't of a quality he wanted to promote...

But I like an alternative explanation: the benchmark is extremely good and extensively used within Anthropic. That's why they're the best at it. And he doesn't want other labs to get the idea to train on it

Locodiff tests something every codegen agent has to do: understand the state of a file after editing it several times

After 6 months Claude code has finally caught up to Mentat :( Luckily we have one feature they'll never have! gpt-5

But seriously it's such a core feature. I can't believe for 6 months it was not possible to run a server in one shell and curl it from another

The labs are most incentivized to produce high quality benchmarks but they aren't incentivized to publish: 1. No one will take "we're the best at our own benchmark" seriously 2. They help other labs improve

One funny episode was when @bio_bootloader published loco diff @alexalbert__ retweeted and then quickly deleted his retweet. No way to know why and probably the best explanation is on reflection he thought the work wasn't of a quality he wanted to promote...

But I like an alternative explanation: the benchmark is extremely good and extensively used within Anthropic. That's why they're the best at it. And he doesn't want other labs to get the idea to train on it

Locodiff tests something every codegen agent has to do: understand the state of a file after editing it several times

But I like an alternative explanation: the benchmark is extremely good and extensively used within Anthropic. That's why they're the best at it. And he doesn't want other labs to get the idea to train on it

Locodiff tests something every codegen agent has to do: understand the state of a file after editing it several times

One funny episode was when @bio_bootloader published loco diff @alexalbert__ retweeted and then quickly deleted his retweet. No way to know why and probably the best explanation is on reflection he thought the work wasn't of a quality he wanted to promote...

But I like an alternative explanation: the benchmark is extremely good and extensively used within Anthropic. That's why they're the best at it. And he doesn't want other labs to get the idea to train on it

Locodiff tests something every codegen agent has to do: understand the state of a file after editing it several times

What is this graph!?

Not only am I not prescient I am not an original thinker. Not my day

Could be 3, 6, 8 or 0. 5? I just don't see it

Tbh 6 would make a lot of sense because 4.5 was originally supposed to be 5 right?

I'm holding strong. No way openai releases something called GPT 5 tomorrow

Should I go all in?

I should read the rules. If they release gpt 5 without the hyphen does it resolve yes?

It makes total sense to me that no one watches the WNBA. What I can't figure out is why people do watch the NBA

Poverty needs no explanation

gpt-5 now live on mentat. It seems to have a pretty different approach from claude. But maybe a good one

Only asked it for one thing so far but I was pretty impressed. 4 commits in 15 minutes for $2.23 getting all tests to pass and making a fairly sizeable PR that worked for me first try

Should I go all in?

I should read the rules. If they release gpt 5 without the hyphen does it resolve yes?

Labs should publish their swe bench unverified performance as a measure of their overfitness

swe bench verified performance is of course also a measure of over fitness but people don't realize

Do any of my followers use and recommend them?

wtf this is a $1600 monitor?!

I know marketing works on me because I see a tweet like this and instantly start thinking of buying an e-ink monitor.

Do any of my followers use and recommend them?

wtf this is a $1600 monitor?!

Which to be fair is probably why we never did well at the actual world finals

But c'mon, you can't figure out how to reverse a linked list from just the problem description?

But okay here's a point on the memorization side: The first words I said at my interview for my current job were "Oh yeah this is Huffman coding I learned it in logic class in 2015"

But c'mon, you can't figure out how to reverse a linked list from just the problem description?

But okay here's a point on the memorization side: The first words I said at my interview for my current job were "Oh yeah this is Huffman coding I learned it in logic class in 2015"

I read the timeless way of building and now I see Alexander posting all the time. Was I just glazing over it before?

It's not subtle either. Every day I see at least one tweet that says "Christopher Alexander spoke of this"

I'm a two time icpc world finalist who considered it borderline cheating to study while I was competing

Which to be fair is probably why we never did well at the actual world finals

But c'mon, you can't figure out how to reverse a linked list from just the problem description?

But okay here's a point on the memorization side: The first words I said at my interview for my current job were "Oh yeah this is Huffman coding I learned it in logic class in 2015"