Alex Wise
@awssnarkitect.bsky.social
Shitposter and lover of safety-critical complex systems
funemployment starts. the future is scary, but we have to celebrate when we can.
youtu.be/F-uQmvoIzW4
youtu.be/F-uQmvoIzW4
The Prisoner (2015 Remaster)
YouTube video by Iron Maiden - Topic
youtu.be
January 31, 2026 at 4:55 PM
funemployment starts. the future is scary, but we have to celebrate when we can.
youtu.be/F-uQmvoIzW4
youtu.be/F-uQmvoIzW4
We used to bankrupt companies trying to get to version 3. Now version three is what makes it to the first PR.
I love that I can build a 6KLoC branch to find out I went the wrong direction and dump it without looking back. That 6KLoC taught me what I needed to learn so the next branch could be a bullseye. This is my primary learning style and it's about 10x faster than ever, good times :)
January 30, 2026 at 5:34 PM
We used to bankrupt companies trying to get to version 3. Now version three is what makes it to the first PR.
IDGI, The 10Q notes: "Approximately 45% of our commercial RPO balance is from OpenAI. The significant remaining balance grew 28%."
That's competitive w/ AWS and GCP growth. There's an actual hyperscaler biz underneath the AI, which is more than many can say.
That's competitive w/ AWS and GCP growth. There's an actual hyperscaler biz underneath the AI, which is more than many can say.
Microsodft dropping -10% because they said 45% of their future cloud business will be from OpenAI and the market assumed they’ll never get the money is a watershed moment for the AI bubble.
January 29, 2026 at 11:33 PM
IDGI, The 10Q notes: "Approximately 45% of our commercial RPO balance is from OpenAI. The significant remaining balance grew 28%."
That's competitive w/ AWS and GCP growth. There's an actual hyperscaler biz underneath the AI, which is more than many can say.
That's competitive w/ AWS and GCP growth. There's an actual hyperscaler biz underneath the AI, which is more than many can say.
How much James Cameron perfectionism energy will it take to fix this? The whole Zemeckisian thesis is that CGI films should mean you make fewer compromises on your vision.
January 29, 2026 at 7:41 PM
How much James Cameron perfectionism energy will it take to fix this? The whole Zemeckisian thesis is that CGI films should mean you make fewer compromises on your vision.
Nobody is beating the wash hire allegations anymore.
I'm hopeful that the market metagame will eventually have to move to discount the value of repeat Q1 layoffs, given that retaining your employees is such an obvious competitive advantage. Might be a long time, though.
I'm hopeful that the market metagame will eventually have to move to discount the value of repeat Q1 layoffs, given that retaining your employees is such an obvious competitive advantage. Might be a long time, though.
January 29, 2026 at 6:12 PM
Nobody is beating the wash hire allegations anymore.
I'm hopeful that the market metagame will eventually have to move to discount the value of repeat Q1 layoffs, given that retaining your employees is such an obvious competitive advantage. Might be a long time, though.
I'm hopeful that the market metagame will eventually have to move to discount the value of repeat Q1 layoffs, given that retaining your employees is such an obvious competitive advantage. Might be a long time, though.
Worried I am being an out-of-touch fuddy-duddy for thinking that Moltbot looks incredibly jank and I don't trust it with my data (but would trust a letta agent I vibe-coded). Anyone have arguments for why Moltbot is worth checking out and recalibrating my vibes on?
January 29, 2026 at 5:32 PM
Worried I am being an out-of-touch fuddy-duddy for thinking that Moltbot looks incredibly jank and I don't trust it with my data (but would trust a letta agent I vibe-coded). Anyone have arguments for why Moltbot is worth checking out and recalibrating my vibes on?
A lot of people clowned on sama for the "Universal Basic Compute" thing, but a world where we subsidize beneficial AI use and tax/punish antisocial use seems prudent and possibly necessary.
It would be great if we could frame the debate as "how do we discourage the uses of AI that are detrimental to society, encourage the positive, and account for externalities?" But the groups with the most power and loudest voices actively don't want that.
January 29, 2026 at 4:42 PM
A lot of people clowned on sama for the "Universal Basic Compute" thing, but a world where we subsidize beneficial AI use and tax/punish antisocial use seems prudent and possibly necessary.
If you aren’t trying to quantify trust on your team you should be. Kirkpatrick-style, by asking the team what the level of trust is.
Trust is interesting because it’s not quantifiable but it certainly is palpable. Teams with trust move lightning speeds faster than those without.
January 29, 2026 at 3:25 PM
If you aren’t trying to quantify trust on your team you should be. Kirkpatrick-style, by asking the team what the level of trust is.
Reposted by Alex Wise
I don’t think there’s a way forward to user-algorithm alignment that doesn’t go through user steering and feedback. Anything else will almost certainly be misaligned with our own values
This is pretty fascinating: academic.oup.com/pnasnexus/ar...
(1) Twitter's algorithm makes its users feel worse about their outgroup
(2) Users do not prefer the tweets it suggests relative to simple baselines
(1) Twitter's algorithm makes its users feel worse about their outgroup
(2) Users do not prefer the tweets it suggests relative to simple baselines
Engagement, user satisfaction, and the amplification of divisive content on social media
Abstract. Social media ranking algorithms typically optimize for users’ revealed preferences, i.e. user engagement such as clicks, shares, and likes. Many
academic.oup.com
January 28, 2026 at 4:07 PM
I don’t think there’s a way forward to user-algorithm alignment that doesn’t go through user steering and feedback. Anything else will almost certainly be misaligned with our own values
Reposted by Alex Wise
Stoked to launch the first @aetheros.computer app from our docs suite today!
PitchDeck is an atproto app that lets you create presentations and share them on the web or in person.
It includes a presenter mode with keyboard/swipe support, a synced laser pointer, speaker notes and a built-in timer.
PitchDeck is an atproto app that lets you create presentations and share them on the web or in person.
It includes a presenter mode with keyboard/swipe support, a synced laser pointer, speaker notes and a built-in timer.
January 27, 2026 at 6:26 PM
Stoked to launch the first @aetheros.computer app from our docs suite today!
PitchDeck is an atproto app that lets you create presentations and share them on the web or in person.
It includes a presenter mode with keyboard/swipe support, a synced laser pointer, speaker notes and a built-in timer.
PitchDeck is an atproto app that lets you create presentations and share them on the web or in person.
It includes a presenter mode with keyboard/swipe support, a synced laser pointer, speaker notes and a built-in timer.
Reposted by Alex Wise
You thought we were done?
No.
Another interop is coming soon.
No.
Another interop is coming soon.
January 27, 2026 at 1:32 AM
You thought we were done?
No.
Another interop is coming soon.
No.
Another interop is coming soon.
This is the future
Yessss YESSSSSSSSS **steeples fingers together**
This is the real promise of the Atmosphere: a seamlessly interconnected network of social apps working together. Look out for Streamplace's integrations with @smokesignal.events and @leaflet.pub, landing soon!
January 27, 2026 at 1:14 AM
This is the future
I gotta do a rewatch to see how this tracks. Honestly would make a lot of sense.
the ship's computer is actually benevolent ASI and handles everything itself, everything it shows the crew on computer screens is just slop to make them feel useful
January 26, 2026 at 12:09 AM
I gotta do a rewatch to see how this tracks. Honestly would make a lot of sense.
Doing my part to end xenophobia by introducing my parents to Butter Chicken.
January 21, 2026 at 5:29 PM
Doing my part to end xenophobia by introducing my parents to Butter Chicken.
Or maybe these leases are so cheap because these apartment buildings tend to churn ownership and evict the old tenants?
You can get a lifetime lease on a nice apartment in the heart of the city for like the cost of like 1k cheap handguns that are sometimes just lying around, so maybe these guys are just doing it for love of the game.
January 20, 2026 at 11:07 PM
Or maybe these leases are so cheap because these apartment buildings tend to churn ownership and evict the old tenants?
Reposted by Alex Wise
I think we’ve got better and more culturally-varied street food than I could have imagined, cool jackets galore, and drugs that make us think we’re really fast and strong before they kill us.
we're getting all the downsides of cyberpunk (social alienation, ruthless hyper capitalism, digital mass surveillance state) but none of the promised upsides (cheap street food, cool jackets, super drugs that make you really fast and strong before they kill you)
January 20, 2026 at 10:02 PM
I think we’ve got better and more culturally-varied street food than I could have imagined, cool jackets galore, and drugs that make us think we’re really fast and strong before they kill us.
Now ask yourself how important it will be that you sprang for the 4K UHD release of the movie instead of the Blu-Ray.
HOLY SHIT, let me show you what i just did
January 20, 2026 at 8:06 PM
Now ask yourself how important it will be that you sprang for the 4K UHD release of the movie instead of the Blu-Ray.
Phone numbers as identity (and access to emergency services) are at the top of my "we're a quarter through 21c, we need to kill off these things from the 20th" list.
Yeah, telcos today do the hard work of verifying ID, but they also spend a lot of effort making sure they are the only ones who can.
Yeah, telcos today do the hard work of verifying ID, but they also spend a lot of effort making sure they are the only ones who can.
interesting discussion subbranch on HN about whether AT is a dead end because client fragmentation makes products uncompetitive. my thoughts:
news.ycombinator.com/item?id=4666...
news.ycombinator.com/item?id=4666...
January 20, 2026 at 6:08 PM
Phone numbers as identity (and access to emergency services) are at the top of my "we're a quarter through 21c, we need to kill off these things from the 20th" list.
Yeah, telcos today do the hard work of verifying ID, but they also spend a lot of effort making sure they are the only ones who can.
Yeah, telcos today do the hard work of verifying ID, but they also spend a lot of effort making sure they are the only ones who can.
In the universe where genderswap ghostbusters was a blockbuster and they started grabbing reels out of the vault to get the same treatment, genderswap animal house would've been the crown jewel.
January 20, 2026 at 4:13 PM
In the universe where genderswap ghostbusters was a blockbuster and they started grabbing reels out of the vault to get the same treatment, genderswap animal house would've been the crown jewel.
Predictions on when G Workspace and O365 bundle a service that builds apps by prompt on their clouds? Would be an absolute game changer for SMB and all the SaaS and consultancies that serve them. (I would still call AI Studio a Cloud/Eng tool).
📄 ✍ Late to the party, but just created a document signing app with Google AI Studio's Build in four iterations!
It asks the user to upload a PDF, then converts each page to an image and uses Gemini to define bounding boxes around open fields that would need to be completed.
It asks the user to upload a PDF, then converts each page to an image and uses Gemini to define bounding boxes around open fields that would need to be completed.
January 20, 2026 at 4:02 PM
Predictions on when G Workspace and O365 bundle a service that builds apps by prompt on their clouds? Would be an absolute game changer for SMB and all the SaaS and consultancies that serve them. (I would still call AI Studio a Cloud/Eng tool).
$5 says the next version schedules doordash to make sure her big man remembers to eat.
This is it. This is The Great Filter. We've created Sexy Barista Is Super Interested in Watching You Solo Game (SBISIIWYS).
January 20, 2026 at 3:34 PM
$5 says the next version schedules doordash to make sure her big man remembers to eat.
we even get public retros when a free, self-hosted feed goes down. Tech culture has grown up a lot in the last ten years.
cause -1 array index access => the program crashes.
Lesson learned for Go - always use len(X) == 0 instead of X == nil because the former supports both nil and empty slices.
Lesson learned for Go - always use len(X) == 0 instead of X == nil because the former supports both nil and empty slices.
January 20, 2026 at 3:05 PM
we even get public retros when a free, self-hosted feed goes down. Tech culture has grown up a lot in the last ten years.
Reposted by Alex Wise
my father teaches computer science and he's describing this variance too: the kids who were flunking before still are now, but the kids who love this stuff are making the best projects that college has ever seen by leaps and bounds
January 19, 2026 at 8:57 PM
my father teaches computer science and he's describing this variance too: the kids who were flunking before still are now, but the kids who love this stuff are making the best projects that college has ever seen by leaps and bounds
I've talked about it other places, but the idea that we're building all these datacenters because we want to train a 60T parameter model or something is just not correct! We're building them because we're forecasting user growth on systems that have dogshit performance for concurrent requests!
FWIW I like the whole of Cory's post, even the claim about horses if you read it narrowly as being about building out datacenters to train bigger and bigger models. But this is not the predominant reason we are seeing datacenter buildouts (this is a common misconception!)
January 18, 2026 at 8:23 PM
I've talked about it other places, but the idea that we're building all these datacenters because we want to train a 60T parameter model or something is just not correct! We're building them because we're forecasting user growth on systems that have dogshit performance for concurrent requests!
This analogy is difficult for me bc he's using all of the nouns used to describe the "humans aren't horses" fallacy, but seems to be making the argument that humans won't need to adapt because the machines won't dislocate tasks currently performed by humans.
Córy Doctorow with another verbal bullseye: pluralistic.net/2026/01/13/n...
January 18, 2026 at 7:15 PM
This analogy is difficult for me bc he's using all of the nouns used to describe the "humans aren't horses" fallacy, but seems to be making the argument that humans won't need to adapt because the machines won't dislocate tasks currently performed by humans.