rem

spent the last couple weeks building my first recsys project. - steam game dataset - two-tower neural net for retrieval - xgboost for ranking - sagemaker for training - fastapi backend - vibe coded frontend to match steam's aesthetic i'm sure i could've squeezed more gains with better features but sagemaker was getting expensive. plan to go deeper into computer vision next so will be posting about that soon.

@juliarturc one of the best ever, just binge other hbo shows

What does one do with their lives after finishing Succession? Nothing will ever rise to that level of writing and acting again. And the freaking soundtrack…

@yeeeeezos there's so much potential with what you can do now. spend a bit of time getting up to speed with skills and subagents and coding will never be the same

@remstack good looks, been a while since i tried claude code, seems like a great time for another go

what's the META in vibe coding these days? not asking about the best model, more about the tooling. the space changes so fast & it's exhausting to keep up tbh. cursor + agent mode? codex? claude code? no gatekeeping man. got the urge to tinker around this weekend

@Temujinsrule @yeeeeezos the other labs target all consumers with their models but anthropic recognised that their main use case was with engineers and went all in on that. in other words, if their models are ever *not* the best for coding, anthropic should be very worried for their future.

@remstack @yeeeeezos Is Claude really that good for coding? What makes it better than the others

@alxfazio do you have any skills or commands that are instrumental to your work other than planning code?

finally carved out some time to finish my claude setup, rounded out all the slash commands and skills. now i can fly through my projects with almost zero prompting. feels nice

@yeeeeezos backprop 🗣️

name something canada has invented? arcteryx aritizia backpropagation basketball blackberry five-pin bowling imax insulin java javascript lululemon poutine slack shopify telephones walkie-talkie you should fear any human that willingly endures -30ºC temps and 6 months of winter

@scaling01 damn wtf

SimpleBench results extremely disappointing for GPT-5.2 GPT-5.2 scores below Sonnet 3.7, an almost 1 year old model GPT-5.2 Pro doesn't fare much better, barely beating GPT-5

@Techniacus gm tech!

Good morning let's finish strong

@k_flowstate Gm flow

Good morning lads

can’t believe i’m sat on a train working on my project with claude code. on my PHONE. how is this possible. honestly generational stuff.

@yeeeeezos can’t be switching every couple weeks man

@remstack my credit card issuer is probably so confused

ffs do i really have to resub to openai again? literally just switched to gemini a couple weeks ago.

@scaling01 hope not

So do I resubscribe? is GPT-5.2 good enough?

@alxfazio does this mean i don’t have to switch my sub? thank god

> opus 4.5 still ahead of gpt-5.2 on webdev arena > gpt-5.2 thinking still weaker than opus 4.5 on tool calling

@AndrewCurran_ damn august 2025 cutoff

GPT-5.2 -400,000 context window -128,000 max output tokens -Aug 31 2025 knowledge cutoff

@compuficial have you tried it out? how is it?

Benchmark performance is not indicative of real-world results.

@AlexHumphries25 looking forward to it bro

We are cooking... All will be revealed with time.

@KevinMDaBess make today great

gm anyone mind if i make today a great day?

@prthkys looks fire

@yeeeeezos i hate seeing openai win

they can't keep getting away with this

@giosaia 100%

Don't make a long list of goals for 2026. We all know you won't complete 99% of the tasks, because you lose motivation. Set 3. Move the needle slightly and you'll make more progress than not moving it at all.