@dprophecyguy @ryanvogel @steipete insanely dumb for 99.99% of people, but considering he effectively has unlimited tokens it's not a bad idea
English
wqite
53 posts
ahhh that’s what @steipete is using so much
he’s running codex on every commit
Greg Brockman@gdb
run codex on every commit
English
@KRR1751 @TechPowerUp Fun and games until you try to use the microsoft store and xbox services. Useless custom OS
English
Microsoft Is Testing a Windows 11 Feature That Maxes Out CPU Speed for Faster App Launches tpu.me/rxj5
English
@RaviTejaKNTS @JasonBotterill then we had a misunderstanding, gpt models are great tools yes
5.4 was just the worst gpt model in a VERY long time other than it's raw speed & efficiency.
very bad personality, often bad coding practices, bad decisions, unable to follow instructions, etc.
5.5 is peak tho
English
@qwqite @JasonBotterill i dont give a damn about benchmarks. i use ai like a tool. GPT models are great tools.
English
@RaviTejaKNTS @JasonBotterill benches? they were good tools before but have you ever even bothered reading how terrible the eval process is ? lol
English
@qwqite @JasonBotterill lol, nah. They are good tools. Not conversational
English
yay another worthless benchmark
John Yang@jyangballin
How much of SQLite, FFmpeg, PHP compiler can LMs code from scratch? Given just an executable and no starter code or internet access. Introducing ProgramBench: 200 rigorous, whole-repo generation tasks where models design, build, and ship a working program end to end. 🧵
English
@realaicoach @DavidKPiano 5.5 is just too fast & has too much baseline performance & consistency for opus 4.7 to matter
opus models have always been inconsistent but if you fought with them, they would easily produce the best results.
it’s still true, but terrible roi vs before
English
@realaicoach @DavidKPiano opus 4.7 is very similar to sonnet 3.7 in the sense that the common consensus is that it’s weird & a downgrade, however it’s definitely an upgrade in many ways
this time around though there exists other better options unlike last time (5.5)
English
@yacineMTB they def increased it insane amounts, at some point on 5.4 on a business account it took 8 prompts with subagents to get rid of my entire weekly usage
5.5 on pro lite it’s been 2 days and i’m down 2%
i assume it’s due to the /goal stuff they added
English
@solxiz @ThourCS2 @TitanHoloCS @WindowsCentral tweakers nowadays dont even pretend to read windows internals, i’m not a random just not a scammer.
English
Italiano
Microsoft is working on a Windows 11 update codenamed Windows K2 with a focus on improving gaming performance and reducing bloatware, according to @WindowsCentral
They are using SteamOS as a benchmark for performance. It will also lower idle memory use and less AI clutter.
English
@solxiz @ThourCS2 @TitanHoloCS @WindowsCentral 99% of tweakers are clueless morons that do not understand anything they are saying & repeating the words of other clueless morons
English
@ThourCS2 @TitanHoloCS @WindowsCentral Not at all. There's always a lof of work to do, no matter how well "optimized" a SO could be.
English
@d4m1n @Zain_Wania i really hope to god for the sake of anyone using it that it’s not worse now than how it was before lol
English
@Zain_Wania nah man it was a lot better ~3mo after it launched with worse models.
It's been bloated and buggy for months now, they are just now starting to fix things.
I've been using CC daily btw pretty much since launch. Still use it
English
English
@NielsRogge Not to defend slop, but ForgeCode and multiple others ranked in the top few were shown to be gaming tb2 using prompt injections / system prompt hints to boost their scores. We still beat CC without needing any gimmicks tho 😼 harness is everything @droid
English
FYI Claude Code is mostly a vibe-coded product (as they say, 100% written by Claude)
It's the worst harness for Opus 4.6 among ANY harness on Terminal-Bench 2
Matt Pocock@mattpocockuk
I feel sorry for Claude Code I know they're not the one. I'm not overcommitting - not investing too hard I wonder if they know I'm pulling away
English
@redtachyon 5.3 - much more efficient than 5.2, better at using the codex harness, very clean direct model
5.4 - useless garbage
5.5 - likely the final model in this mini family, seems to be pretty competent & acts a lot more like an anthropic model compared to 5.0
English
3.5 - silly, interesting, largely useless
4 - first actually useful model, at least on some things
4o - multimodal, misaligned, oneshot normies
4.5 - bigger, more raw, very interesting
o1 - first reasoner, impressive for its time
o3 - absolute beast, still an incredible model
5 - o3 in a trenchcoat
5.1 - people were mad at 5 so this was a bit better
5.2 - codex era, great agentic performance
5.3 - ?
5.4 - ??
5.5 - ???
English
@rileybrown It's not that crazy... they steal/train on the actual images that include barcodes lol
English
Serious question. Has anyone ever noticed meaningful regressions in Codex/OpenAI models?
I feel like we talk about this a lot w/ Anthropic but I've never seen a similar discussion with OAI.
English
@bcherny @realsigridjin Why is 4.6 not accessible @bcherny ? Even OpenAI rolled back that bad policy. Old models have to be supported for a period at least
English
in claude web
- opus 4.7: only adaptive thinking mode
- opus 4.6: i can turn on/off reasoning mode
so basically you can't control thinking mode
English
@mahm_darwish @KhalidWarsa you’re ignoring the implications of caching very hard
English
Tbh I am not sure how this is a good argument, using only Claude code I am consuming $3-4k worth of tokens monthly.
I don’t care about OpenClaw, but I want to use the cli mode for my software or any software that simply uses Claude code.
Automation is a normal progression for LLM tools.
English
I’m with Anthropic for banning OpenClaw cause you can’t consume $2,000 worth of tokens on a $200 subscription and expect Anthropic to foot the bill.
Your “24/7 running agents that produce slop and pollute the internet” are net negative on society. Pay actual usage or gtfoh.
English