
Executive Producer https://teapawt.net/t.co/88mStAaGOe | @getsentry Parenting Podcast - https://teapawt.net/t.co/6XrVfZWdbE Previously Creator of @leveluptuts
Colorado
Joined January 2009
- Tweets25.8K
- Following2.5K
- Followers65.1K
- Likes90.5K
Pinned Tweet
This is the most important video I've ever made.
In this, I explore the negative and positive mental health effects of AI coding, with data and expert commentary.
youtu.be/iPUn1Fnfn0k
This is my experience too
Wow.
Neon is a great model in exactly the ways that GPT 6.2 is terrible. Let me explain what I mean.
One thing I love about Model Zen Garden, is amount of subjectivity in the result. Yes we rank things on blind comparisons and you can see which models do better and worse, but actually walking around the garden gives you the most visceral feel for where a given model is “jagged”.
You see, jaggedness isn’t always a negative. There are some models, for example, that score very low on the blind tests because they have poor textures and 3d mechanics, but are actually great models for work - you can easily see this in model zen garden because interactions “work” even though the world may look bland.
The GPT line of models have long been this way, the world they make is small, boring, and bland, but nearly everything works perfectly. Interestingly, Astra 6.2 is the first GPT model where the textures and 3d models are really jaw-dropping. However, its the first recent GPT model that is has left bigs gaps in the way things are actually “put together”.
This is a major step-back for OpenAI. The one thing I could always count on, really since GPT 5.5, is that their models were fastidious and didn’t leave obvious holes in logic or execution. Unless you tortured the model (ultra mode) you could trust it. And trust is everything. You cannot trust 6.2.
By these same subjective metrics, Neon is one of the most trustworthy models I’ve ever seen. It achieves this by being the “smoothest” model yet (only Opus 5.5 is close). There are virtually no holes at all. It’s slightly above average in every single metric: cost, speed, token efficiency - all very middle of the road. Except one thing: it gets EVERYTHING right.
You can see and feel this in zen garden. It’s garden isn’t the most beautiful necessarily, but its very good, its bridge is actually connected to the ground (rare), the mechanics for feeding the fish are perfect - the food has momentum and proper gravity, leaves ripples in the water, and the fish actually eat individual food pieces properly…it works really really well. Teapots tip the right way (rare), and the rake actually has pitch and yaw. The bell chime actually has to hit the bell to make sound and it travels the right direction (rare) - it even rings on the rebound hit.
The end result is a model that scored nearly as well as Fable 5.2, for 1/4 the price - and honestly, got its interactions even more “right” than Fable.
In “real-world” use we are seeing the same thing, consistent execution, a smooth trust profile, and dogged tenacity.
As my trust grows with it, I’ll give it increasingly responsibility and will be sure to let everyone know where we start to see cracks, but for now, it looks like you need to get another subscription.
3
35
6,532
ah yes, i've always heard of the latest mathematics hypothesis that's been solved and totally understand it. quite exciting. i've been waiting for this moment all my life.
6
53
3,642
Haven’t finished your holiday shopping yet? The syntax team is back with huge gift guide for tech people.
Remember circling everything you wanted for Christmas in a holiday catalog?
We made the Syntax WishBook™️ with all of our favorite things and sick picks from this year's gift guide.
Circle items, share the link, or print an order form for Santa.
wishbook.syntax.fm
chat, is this legit?
We shrank 1 GB of Wikipedia to 105.39 MB, losslessly.
Introducing tufazip: the best entry on the Large Text Compression Benchmark using a neural network as its sole predictor. Everything above it mixes a transformer with hand-built models.
Full post 👇
tufalabs.ai/research/tufazip
10
61
14,983
i'm looking to interview some AI benchmark experts, who should i talk to?
@arena @datacurve
where my agency folks at? most of my career was rapidly building sites for clients, those clients are going to need agents
Everyone who works at an agency that builds software (apps/websites etc) for clients has starting getting requests from their customers for custom agents.
If that's you, Standard Agents is literally a life line for your business. Send me a DM asap.
7
21
4,880
Scott Tolinski retweeted
"The way you make your full code observable is by not having to think about telemetry at all" – @MichaelArnaldi
Especially true when paired with:
"Discipline doesn't scale" – @dillon_mulroy
Watch this @syntaxfm:
youtube.com/watch?v=fLi2Hivl…
Ben is one of those people that really gets it. I love what they are doing over here.
We messed up.
We built an AI PM, but nobody wants another agent. They want to use their own.
So we rebuilt the platform.
Introducing Modem 2.0: finds problems buried in customer channels, tracks them, and follows up when fixes ship. Works with the agents you use every day.
2
80
8,081
Talk not live yet, based on this video youtube.com/watch?v=iPUn1Fnf…
1
2
801
Scott Tolinski retweeted
Replying to @catalinmpit
@stolinski has a great video about this youtube.com/watch?v=iPUn1Fnf…
Scott Tolinski retweeted
If you haven’t yet built a system for shipping to prod without human code review, then…with love…maybe software isn’t the thing for you.
live now
We're live to chat about expo/exact which lets you write once run everywhere, Fable 5.2 is here (it uses a lot of tokens), iOS nerfs autocomplete and more!
See you in the chat.
youtube.com/watch?v=rVgRWm59…
4
2,081
Write Once, Deploy Everywhere ⟡ Fable 5.2 Token Hungry ⟡ iOS Autocomplete Nerfed ⌁ Syntax Weekly ⌁ teapawt.net/i/broadcasts/2JQeqdzzg…
We're also live here on @syntaxfm
Is the dream of write once deploy anywhere actually here?
youtube.com/watch?v=rVgRWm59…
9
2,081
Is the dream of write once deploy anywhere actually here?
youtube.com/watch?v=rVgRWm59…








