@tech_optimist
Account based inCanada
About this account
- Account based in
- Canada
- Connected via
- Canada App Store
Account-level information from X, not a live location or the device used for a specific post.
Founding AI engineer & researcher @hdclabs_ai Blogging @ https://nitter.cf/t.co/gLektr01zQ and https://nitter.cf/t.co/rU0gaNPIgm
Toronto 🇨🇦
Joined October 2013
- Tweets4.4K
- Following1.9K
- Followers2.4K
- Likes16.6K
Sadly, there will be a large fraction of human society that goes this way. Just like many don't go to the gym to stay physically fit, the same will happen with excessive infiltration of AI into our daily lives. The brain can atrophy if it's not made to "sweat".
Let's all train those critical thinking muscles every day!
So excited to try this out! Love the new releases of DSPy, soooo much going on that's making me wanna use it every day
Sometimes, being at the frontier of using these models is exhausting. Being on here is exhausting. Everyone and their mother has an opinion on why model X is great at this, but not that, blah blah.
Reminder to self. Get outside, touch grass, and remember that this is an echo chamber with a faulty algo and that reality is subjective. It's all in the mind. 😁 🧘🏽♂️🧘🏽♂️🧘🏽♂️
Okay, enough is enough. Opus 5.5 users. TRY WRITING ANYTHING OPEN-ENDED WITH IT. Not code, not docs. It truly sucks in style. None of my past skills work with it. I'd love to learn how anyone has gotten it to write something that's not code/docs with an ounce of personality. It looks, feels and sounds like a goddamn robot when it writes (which it is, but still).
Okay, from what I gather across the interwebz, both OpenAI and Anthropic are serious about improving the writing quality of the models, if the GPT-6 Sol and Opus 5.5 releases are anything to go by.
Models that write more like humans are, after all, more in tune with the humans who consume what they produce. Reallly curious what post-training tricks have been used to get the models to write the way they do now!
Did I just hear they're better at WRITING 😍😍
GPT-6 Sol and Luna are out. Not only are they a very significant improvement across the board, but also in writing and general "you know when you try it" quality.
We are also permanently reducing the API price by 50% making both of them viable for a ton of new usecases and making your usage go further too, even on the subscriptions.
And one more thing. We are loading a banked reset into all accounts of our Plus, Pro and Business users. Let's go!
openai.com/index/introducing…
One of the largest Monarch butterfly migrations in recent times is currently underway in North America.
Apparently, a "perfect storm" of breeding conditions have made this one of the largest populations of butterflies, and I've definitely been seeing these little critters everywhere in Toronto!
It's quite remarkable how tough and hardy they must be to make that long a migration down south.
Although I don't use the actual image below in the post, I couldn't resist using this example to illustrate the points we're making about HDC and hypervector dimensionality in general － it definitely fits the season we're in right now!
Post: hyperdimensionalcomputing.ai…
In HDC, it's natural to wonder, why would we compress an image's high-dimensional representation, only to then expand and then spread the result across 10,000 dimensions?
The reason is quite illuminating. The first step (compression) in HDC decides which differences matter for the task. The second (expansion) gives the retained features a wide and distributed code to compute with. Both matter.
Let's understand this better using an analogy, comparing a photo taken in sunlight, and the same photo taken in the shade.
1/6
Also, TIL, there exists a "Monarch highway" that restores the habitat for these butterflies as they make that long migration down south. Just like we humans make rest stops, the butterflies need a good place to replenish too! Fascinating.
npr.org/2026/09/20/nx-s1-597…
Prashanth Rao retweeted
Hating Pearson is an important part of Canadian identity, like hating Air Canada even though it’s sort of okay.
Prashanth Rao retweeted
If Jev had launched 1-2 years ago, despite being a pareto frontier classifier, it would have flown under the radar. The fact that it arrived right as so many of us are feeling extreme pain from token costs is why there's so much hype. It solves a real immediate pain point.
AGI mathematicians are here to stay 🤯
Also, wth can somebody explain to me how the heck Terence Tao isn't on that advisory committee?!!
We’re working with an independent advisory group of mathematicians to help OpenAI responsibly share advances in AI and mathematics.
The group will advise on how we assess and communicate new mathematical results, uphold academic and professional standards, and build tools that support mathematical research and learning.
Through this work, we want mathematicians to be at the center of shaping how AI supports mathematical understanding and how its benefits reach the wider community.
openai.com/index/advisory-gr…
Prashanth Rao retweeted
I didn't start using agents in any meaningful way until December 2025. I got full-on psychosis over the holidays.
We had Opus 4.5 and GPT-5.2 (next up was GPT-5.3-Codex). Insane that it's been less than a year.
Love this analogy. Sounds a lot like my experience bashing my head against the wall with these models when writing anything of value these days.
my quick and dirty analogy for the model's flavor of jaggedness in 2026 is kind of like...
if you ask for a taxi ride but someone hands you a catapult that is highly pre-aimed on a very small number of destinations in advance
initially, you're super impressed by the speed and distance of transportation, but you can never quite use it to land at your actual freaking destination, no matter how many times you're willing to get thrown flying through the air :/
excuse the dumb analogy but that's the informal version of the intuition
Indeed. Jev has kickstarted an explosive growth cycle in discriminative models in open source, that will start to form the backbone of more and more harnesses. I don't think the ecosystem will ever go back to using generative models for routing ever again.
thoughts on Typesafe/Jev: I'm surprised that a general-purpose classifier (or discriminative model) can be just as interesting to the public as a general-purpose generative model. I've worked on search and discriminative models for years, so maybe I should have called this upfront, tbf I never thought about making them general-purpose beyond search. Typesafe also positions Jev really well as "System 1", a complement to the generative models we already have. Humble, unassuming, and sidestepping the frontier-lab warzone.
Jev's success could be disruptive to today's agentic systems in many ways, such as tool-calling and routing patterns. Over the last three years the consensus was to use a small generative model for routing, tool calling, MCP etc. Time to pause and rethink such architecture, because we'll probably see tool calling and routing move back to discriminative models. Whoever ships the next Jev-like open-weight base model will likely win the community.
Prashanth Rao retweeted
A graph tells us how records connect. Could a fragment of that connected structure become a search query, even when names or details are missing? Yesterday's post on delayed-order example posed such a question. Today, let's see one way to build that search from a small graph.
hyperdimensionalcomputing.ai…
1/7
With all the (rightful) buzz around @typesafeai and Jev today, it never fails to blow my mind, how many problems can simply be reframed as "classification problems."
It's no wonder everyone is going crazy over it. This is exactly what the AI world needs.