I used @cerebras code plans for a long while. Even on a “dumber” model the speed gains are totally worth it.
Hard to go back to slow frontier models after tasting 600 tokens/second (still twice as fast as ultrafast 🙃)
Just did one of the playground tutorials in #Jev. It’s everything I could have hoped for. Looking forward to diving deeper. @typesafeai is the answer to something I’ve wanted for a while now 😄
@justKDeng I want to build a coding harness where Jev acts as a supervisor of an LLM. Essentially validating tool calls, making sure the model stays on track, doesn’t fall into loops, etc.
(I’m not able to DM my email to you though due to not being verified)
Jev is the answer to something I've wanted for a while, but as far as LLMs go... I feaking love GLM-5.3-Flash (formerly Ox Alpha)
"I'm the MASTER! Whup whup."
"Ok, i'm just a backup, great."
"Fault, what ?"
It's got a sort of whimsy about it 🤭
I had a @RustFSofficial question and @OpenAI Astra started writing Python to do checks against the actual source code in GitHub rather than search documentation.
Sorry to add more workload to GitHub but that was pretty cool…
A quick look into remote persistent sessions with Superlogical! Not just remote sessions, but some functionality like remote directory lists, too. Superlogical can actually be a full SSH replacement (I don't run SSH on my servers anymore), and I talk about that briefly, too.
This is what makes software delightful and I love it.
Even on Superlogical’s website there’s a “We’re hiring” that animates an SSH command to run and get to the Job Descriptions. Details like that are wonderful ���️
One of my favourite things about building @superlogical is that we all care, a disproportionate amount, about the little conveniences, performance gains and UI treats that most rational people just don’t.
I'm currently working on our scrollback search UI. You basically have two common choices as a designer:
a) insert a toolbar into the view or
b) float a search bar over the view
For terminals, any intrinsic size change causes a reflow - which is jarring for the user to see and potentially pushes offscreen something you were looking at. This means option A sucks.
For option B though, the problem is that you’re potentially obscuring matched words that fall under the search UI since it floats over them. (Every app I tried makes this mistake which is kinda nuts)
So what's the solution? Give our search UI a gentle nudge out the way when it's over something you're trying to see.
Regarding #PhysicalGamesMatter, one critical thing I use a PS5 for these days is playing 4K Blu-Ray movies. If Sony drops support for physical games, I assume by implication they are also dropping support for physical movies. Assuming so, I would never upgrade to a PS6.
Token Maxing is popular because LLMs are slow. You can at least kick off work in parallel. When inference is fast, there’s no time to context switch and your feedback loop goes back to something more akin to pair programming.
"Post-acquisition, the fee stays inside the house. Stripe did not overpay for a middleware company. It swapped a payment processor's take rate for a routing platform's take rate, on a transaction base that tripled in months and has no ceiling."
I partially agree, parents should not blindly listen to their doctor but take their recommendations under advisement. If their doctor is good, they won’t listen blindly to Government recommendations one way or the other.
Not sure why we’re upset with this 🤔
PARENTS: listen to your doctor NOT Trump or RFK Jr.
IGNORE Trump’s Executive Order. There is NO link between vaccines and autism.
In the middle of actual public health crises America deserves leaders informed by actual science. FIRE RFK Jr.