Allen Pike
@allenpike.com
Building Forestwalk.ai, hosting Infer and It Shipped That Way, and writing about software and other tricky things. ⭐️
Vancouver • @apike elsewhere • allenpike.com
Vancouver • @apike elsewhere • allenpike.com
Finally tried gpt-5-mini on some tasks, hoping at "minimal" reasoning it would be faster than claude-haiku-4.5 or gpt-4.1-mini.
Nope. Two other teams said 5-mini is so slow and unreliable that they can't use it either. Apparently "4 lightning bolts" means "takes 5-10s to process 200 tokens" 💫
Nope. Two other teams said 5-mini is so slow and unreliable that they can't use it either. Apparently "4 lightning bolts" means "takes 5-10s to process 200 tokens" 💫
November 6, 2025 at 11:05 PM
Finally tried gpt-5-mini on some tasks, hoping at "minimal" reasoning it would be faster than claude-haiku-4.5 or gpt-4.1-mini.
Nope. Two other teams said 5-mini is so slow and unreliable that they can't use it either. Apparently "4 lightning bolts" means "takes 5-10s to process 200 tokens" 💫
Nope. Two other teams said 5-mini is so slow and unreliable that they can't use it either. Apparently "4 lightning bolts" means "takes 5-10s to process 200 tokens" 💫
While the AWS team has resolved the severe us-east-1 outages, reading this post-mortem makes one thing clear: they are still suffering from a critical shortage of paragraph breaks
October 23, 2025 at 5:49 PM
While the AWS team has resolved the severe us-east-1 outages, reading this post-mortem makes one thing clear: they are still suffering from a critical shortage of paragraph breaks
If you're getting weird behaviour from ChatGPT 5, check "What traits should ChatGPT have?" under Customize ChatGPT. I had this prompt there to help 4o be more critical, but they trained GPT-5 to be more critical by default, so the prompt pushed it into overcaffeinated paranoia.
August 7, 2025 at 11:19 PM
If you're getting weird behaviour from ChatGPT 5, check "What traits should ChatGPT have?" under Customize ChatGPT. I had this prompt there to help 4o be more critical, but they trained GPT-5 to be more critical by default, so the prompt pushed it into overcaffeinated paranoia.
I almost deleted this notification without thinking, but stopped to read it in a different light. When you have a lot left to do, it's easy to miss the little moments that represent that you're making progress.
So here's to that. 🍻
So here's to that. 🍻
June 25, 2025 at 5:12 PM
I almost deleted this notification without thinking, but stopped to read it in a different light. When you have a lot left to do, it's easy to miss the little moments that represent that you're making progress.
So here's to that. 🍻
So here's to that. 🍻
Libraries often have private, internal API surfaces that must not be depended on. It is important to name these dangerous APIs in an alarming way. This ensures that any consumers of your library will be embarrassed when they inevitably use them anyway
May 22, 2025 at 3:03 PM
Libraries often have private, internal API surfaces that must not be depended on. It is important to name these dangerous APIs in an alarming way. This ensures that any consumers of your library will be embarrassed when they inevitably use them anyway
In two weeks OpenAI has added Instacart's CEO, the Windsurf staff, and Sir Jony Ive himself to their team.
If they can continue this pace, OpenAI may eventually amass the talent necessary to overcome the ultimate product challenge: designing a reasonable model selection UI
If they can continue this pace, OpenAI may eventually amass the talent necessary to overcome the ultimate product challenge: designing a reasonable model selection UI
May 21, 2025 at 7:27 PM
In two weeks OpenAI has added Instacart's CEO, the Windsurf staff, and Sir Jony Ive himself to their team.
If they can continue this pace, OpenAI may eventually amass the talent necessary to overcome the ultimate product challenge: designing a reasonable model selection UI
If they can continue this pace, OpenAI may eventually amass the talent necessary to overcome the ultimate product challenge: designing a reasonable model selection UI
I was invited to share what I've been learning about UX patterns for non-chat AI products at the Vancouver AI Meetup last week, and the video is already up ⭐️ www.youtube.com/watch?v=mRqB...
May 5, 2025 at 3:57 PM
I was invited to share what I've been learning about UX patterns for non-chat AI products at the Vancouver AI Meetup last week, and the video is already up ⭐️ www.youtube.com/watch?v=mRqB...
The Periodic Table of Elements, according to GPT-4o's new, smarter image generation
March 25, 2025 at 7:53 PM
The Periodic Table of Elements, according to GPT-4o's new, smarter image generation
Local chain Save-On-Foods has new in-store flyers promoting their Canadian ownership. Where else can you fill your cart with pumpkin-sized apples, giant jars of demented pasta sauce, lettuce bigger than your cart, and a bunch of mysterious crimson… mush? And don't forget their patented “BEEFY DAIRY”
February 13, 2025 at 3:35 AM
Local chain Save-On-Foods has new in-store flyers promoting their Canadian ownership. Where else can you fill your cart with pumpkin-sized apples, giant jars of demented pasta sauce, lettuce bigger than your cart, and a bunch of mysterious crimson… mush? And don't forget their patented “BEEFY DAIRY”
ChatGPT advanced voice mode is helpful for talking you through stretches! It can suggest different positions, explain the muscles involved, and give tips on technique. Just be aware that if have it describe lying down or getting on your knees, the response might be deemed too spicy
January 28, 2025 at 8:42 PM
ChatGPT advanced voice mode is helpful for talking you through stretches! It can suggest different positions, explain the muscles involved, and give tips on technique. Just be aware that if have it describe lying down or getting on your knees, the response might be deemed too spicy
On a day to day basis, founders typically focus on what's immediately ahead – the daily process of moving the business forward.
This can make it easy to lose sight of what you've achieved. For example, according to LinkedIn, we have achieved a YoY growth rate exceeding two billion percent
This can make it easy to lose sight of what you've achieved. For example, according to LinkedIn, we have achieved a YoY growth rate exceeding two billion percent
January 16, 2025 at 8:34 PM
On a day to day basis, founders typically focus on what's immediately ahead – the daily process of moving the business forward.
This can make it easy to lose sight of what you've achieved. For example, according to LinkedIn, we have achieved a YoY growth rate exceeding two billion percent
This can make it easy to lose sight of what you've achieved. For example, according to LinkedIn, we have achieved a YoY growth rate exceeding two billion percent
It's a new year, and you know what that means: time to confirm your website is still readable in Netscape® Navigator 4
January 2, 2025 at 10:00 PM
It's a new year, and you know what that means: time to confirm your website is still readable in Netscape® Navigator 4
While we'll need to wait until Feb to see how much better OpenAI o3 will do on subjective tasks, it's interesting to see this huge jump on performance on the ARC-AGI benchmark (akin to an IQ test). It's a good sign when a benchmark's originator announces a new, harder benchmark along with your score
December 20, 2024 at 7:57 PM
While we'll need to wait until Feb to see how much better OpenAI o3 will do on subjective tasks, it's interesting to see this huge jump on performance on the ARC-AGI benchmark (akin to an IQ test). It's a good sign when a benchmark's originator announces a new, harder benchmark along with your score
Useful point from @guo.bsky.social at Latent Space Live: don't underestimate unmet demand for better tools.
Most people can't use, say, Photoshop, so the market for creative tools can seem capped. But 10x how many people can do creative work, and latent demand appears.
Most people can't use, say, Photoshop, so the market for creative tools can seem capped. But 10x how many people can do creative work, and latent demand appears.
December 11, 2024 at 6:47 PM
Useful point from @guo.bsky.social at Latent Space Live: don't underestimate unmet demand for better tools.
Most people can't use, say, Photoshop, so the market for creative tools can seem capped. But 10x how many people can do creative work, and latent demand appears.
Most people can't use, say, Photoshop, so the market for creative tools can seem capped. But 10x how many people can do creative work, and latent demand appears.
On the limits of phone-based LLMs, this is a fun passage from @simonwillison.net about running, for the first time, a GPT-4-class model locally on a Mac. (For context, the Apple Intelligence models use roughly 3GB.) https://simonwillison.net/2024/Dec/9/llama-33-70b/
December 9, 2024 at 9:43 PM
On the limits of phone-based LLMs, this is a fun passage from @simonwillison.net about running, for the first time, a GPT-4-class model locally on a Mac. (For context, the Apple Intelligence models use roughly 3GB.) https://simonwillison.net/2024/Dec/9/llama-33-70b/
The modern web stack is amazing – in the way that a giraffe is. As it evolved from simplicity to majesty, some details became deranged, like how its vocal nerve goes all the way down the neck, under the heart, then back up again. But as a whole system, it works, and it's miraculous.
Giraffes, man.
Giraffes, man.
November 28, 2024 at 3:00 AM
The modern web stack is amazing – in the way that a giraffe is. As it evolved from simplicity to majesty, some details became deranged, like how its vocal nerve goes all the way down the neck, under the heart, then back up again. But as a whole system, it works, and it's miraculous.
Giraffes, man.
Giraffes, man.
It's fascinating how with each iteration of advanced aerospace, math, and manufacturing, the design slowly converges to “make a peregrine falcon”
(Bird not to scale)
(Bird not to scale)
November 23, 2024 at 7:41 PM
It's fascinating how with each iteration of advanced aerospace, math, and manufacturing, the design slowly converges to “make a peregrine falcon”
(Bird not to scale)
(Bird not to scale)
I asked Cursor this of an active file as a joke, and the answer was... great
November 18, 2024 at 2:17 AM
I asked Cursor this of an active file as a joke, and the answer was... great
Xcode's generative tooling for Swift might be anemic compared to what Cursor can do for Python or TypeScript, but at least it's entertaining
November 18, 2024 at 2:08 AM
Xcode's generative tooling for Swift might be anemic compared to what Cursor can do for Python or TypeScript, but at least it's entertaining
A friend said iOS Image Playgrounds are an upgrade over Memoji, which is true. I wanted to convey “🤔 hmm”, and that is indeed the feeling evoked
November 15, 2024 at 10:16 PM
A friend said iOS Image Playgrounds are an upgrade over Memoji, which is true. I wanted to convey “🤔 hmm”, and that is indeed the feeling evoked
Apple Intelligence’s Image Playground “Illustration” style in iOS 18.2b is a wild journey
November 13, 2024 at 7:38 PM
Apple Intelligence’s Image Playground “Illustration” style in iOS 18.2b is a wild journey
Every line of code you ship eventually needs maintenance. Assumptions break, the world evolves, users change.
Recently, this would have been a useful hint: “Are you REALLY from Anguilla and not Austria?” Now it's a tiny warning flag: “This product might be under-maintained.”
Recently, this would have been a useful hint: “Are you REALLY from Anguilla and not Austria?” Now it's a tiny warning flag: “This product might be under-maintained.”
November 8, 2024 at 8:10 PM
Every line of code you ship eventually needs maintenance. Assumptions break, the world evolves, users change.
Recently, this would have been a useful hint: “Are you REALLY from Anguilla and not Austria?” Now it's a tiny warning flag: “This product might be under-maintained.”
Recently, this would have been a useful hint: “Are you REALLY from Anguilla and not Austria?” Now it's a tiny warning flag: “This product might be under-maintained.”
While it may never be achieved, by gathering the brightest minds of our generation, frontier AI labs hope to one day crack the monumental problem of naming new models
"After extensive analysis of what we should call the model after 3.5, we have decided to go with 3.5 (new)."
"After extensive analysis of what we should call the model after 3.5, we have decided to go with 3.5 (new)."
October 22, 2024 at 5:25 PM
While it may never be achieved, by gathering the brightest minds of our generation, frontier AI labs hope to one day crack the monumental problem of naming new models
"After extensive analysis of what we should call the model after 3.5, we have decided to go with 3.5 (new)."
"After extensive analysis of what we should call the model after 3.5, we have decided to go with 3.5 (new)."
Every news app should have Artifact’s “Mark as Clickbait” feature.
May 25, 2023 at 4:02 AM
Every news app should have Artifact’s “Mark as Clickbait” feature.
“This church you’re starting, how holy is it going to be?”
“Nothing too unusual, I was just thinking it would be a
“Nothing too unusual, I was just thinking it would be a
May 22, 2023 at 11:58 PM
“This church you’re starting, how holy is it going to be?”
“Nothing too unusual, I was just thinking it would be a
“Nothing too unusual, I was just thinking it would be a