Kuumin

malaysian engineering gremlin 🇲🇾 | EE @ UM | Rust, infra, agents, automation, deep tech | building things until malaysia's gdp notices me

[Kuumin @KuuminDa]

@TokenGremlin Following each reset, the usage limit is reset for EVERY user. Since they're gaining users, it probably gives them a lot of insight at how much compute margin they have before maxxing out their resources, so that they can allocate resources appropriately (2/2)

[Kuumin @KuuminDa]

@TokenGremlin The difference is in sustained compute and peak compute. They can sustain compute very easily, which is why they keep resetting, but they can't handle large concurrent development sessions from Plus and Pro all at once.

I have a crackpot theory, another factor is timing (1/2)

[Kuumin @KuuminDa]

@kunchenguid I've been using Oh My Pi now, and honestly, it started to become daunting. Omp started clashing when I wanted to add pi-dictation, and afterwards it started to feel like it was no longer my harness. Eventually, I switched back to pi

[Kuumin @KuuminDa]

Switching back to pi, oh-my-pi's batteries includedness was nice but eventually the extensibility of it and the eye garbling decos started to irritate me

[Kuumin @KuuminDa]

X Algo, give me more engineering college student tweets

[Kuumin @KuuminDa]

Another Reset?!? Daddy Tibs you're spoiling us when it's not Christmas!

[Tibo @thsottiaux]

This celebration is moved to tomorrow as the button was already pressed today.

[Tibo @thsottiaux]

This celebration is moved to tomorrow as the button was already pressed today.

[Kuumin @KuuminDa]

I feel that this is quite a reasonable move? Even though 5% of the traffic doesn't sound like a lot, 5% is still a lot.

Plus, Elon, whose affliated with Cursor now, wants to sue and bleed OpenAI dry. It's strange for people to complain about not wanting to be associated?

[OpenAI @OpenAI]

We’re ending our partnership with Cursor following its acquisition by SpaceX. Under our proposal, Cursor’s direct access to our models would end on November 12.

We know that the people most affected by this decision are the developers who rely on OpenAI models in Cursor. We care about their experience in this transition and we’re ready to go above and beyond to support them.

https://t.co/OzuCTzUjfX

[Kuumin @KuuminDa]

@ishuagra02 Baffling how Anthropic is running themselves into the ground, while OpenAI is providing food for the starved

[Kuumin @KuuminDa]

@Eri90091Matthew @kimmonismus Considering the service OpenAI does for the community, that minor infraction is worth all the doomhype

[Kuumin @KuuminDa]

YESSS YOU'RE GOOD... KEEP DRAWING PLEASE... you have talent... PLEASE DONT STOOOOP 😭😭

[Anthonny Ng Hon Wei @AnthonnyNgHonwe]

@samgladiator I think my is bad... But I work very hard on this

[Kuumin @KuuminDa]

OH MY GOODNESS AI HAS CAUSED MY BRAIN TO ATROPHY SO HARD... IT'S DIFFICULT FOR ME TO IMPLEMENT STUFF IN RUST NOW WITHOUT AI RAHHHHH! This is the circuit analysis project, btw

[Kuumin @KuuminDa]

@sahilypatel I'm betting on OpenAI. Not because I have any valid reasons to, but because I just don't like Anthropic

[Kuumin @KuuminDa]

@blueemi99 I was thinking about this as well, but honestly I feel like we're not seeing the full picture here.

China matches up on the CONSUMER side, not internal systems-wise. Though, China is doing a good job at pushing OpenAI and Anthropic into pushing better frontier models

[Kuumin @KuuminDa]

@bindureddy @SahilPanhotra I'd like to argue that they wanted to stir up the pot for Gemini to gain more brainshare before they drop a Gemini bomb

or they were just having fun and we're reading too much into it

[Kuumin @KuuminDa]

@januarycomputer This was actually the one of the first AI generated video that I thought was actually pretty good. Alongside the tiger Ben 10 one as well. It still has that distinct AI flavour, but it isn't as bad

[Kuumin @KuuminDa]

@Its_lakshya_ai I saw this on my usage limit, I wonder if it's the same thing.

Also, wasn't this patched because of @theo and @maria_rcks? (Not to shame them, but I would've done the same mistake as well not knowing the consequences)

[Kuumin @KuuminDa]

@KentonVarda I found that rather than asking it to find bugs, it was actually better to ask it to trace the flow of a specific function, and to look for potential bugs that results from the implementation of it.

It constrains the model's view on a specific flow and it becomes more accurate

[Kuumin @KuuminDa]

@theo Honestly, even if it's not the cheapest, the performance to price ratio is really good. I wonder what special sauce they've added to their pre-train or post-train? We really thought it was a big param model.

[Kuumin @KuuminDa]

Also, I realised how people lose their minds at the moment they see the word "train", "model" or "AI" together, even when it's regular NLPs.

It makes sense, when hearing those words together, it's often conflated with transformer and LLMs.