Umer Farooq

1K posts

Umer Farooq

Umer Farooq

@Dev__F

AI Engineer • Building AI that actually works! https://t.co/ctLLJKKaZs

Se unió Ağustos 2023
384 Siguiendo142 Seguidores
Umer Farooq
Umer Farooq@Dev__F·
@gmi_cloud just got day-0 access to nvidia nemotron 3 ultra 550B params, 55B active, 1M token context, hybrid Mamba-Transformer MoE running on H200 and Blackwell infra when your inference provider has frontier models before your internal team does
Umer Farooq tweet media
English
0
0
0
17
Abdullah Kun
Abdullah Kun@bin_zeshan·
Claude extension nay mera browser bohat slow kardia hai
English
2
0
4
469
Umer Farooq
Umer Farooq@Dev__F·
@daniel_mac8 That "not moving everything back from Codex" line hits home. I've been feeling the same split, Opus is great for creative stuff but Codex still wins for the technical heavy lifting.
English
0
0
0
79
Dan McAteer
Dan McAteer@daniel_mac8·
Opus 4.8 is a good model. Tangibly better than Opus 4.7. But not "I'm moving everything back from Codex to Claude" good. For that, I think I'll need to wait for Mythos. I'll prob stick with Codex/GPT-5.5 for coding + technical work and use Opus 4.8 for more creative stuff.
English
31
6
161
12.5K
Umer Farooq
Umer Farooq@Dev__F·
@iruletheworldmo ChatGPT 5.5 hasn't given me a reason to leave tbh. The "switch back to opus" impulse faded once the gap got this wide.
English
0
0
0
2.3K
🍓🍓🍓
🍓🍓🍓@iruletheworldmo·
so what’s the vibe for opus 4.8 vs chatgpt 5.5? seems like there’s no need to switch back to opus? roon has the mandate still?
English
104
5
342
68.9K
Umer Farooq
Umer Farooq@Dev__F·
@danshipper Damn that's brutal. Being sick is the absolute worst, hope you're taking it easy.
English
0
0
0
225
Umer Farooq
Umer Farooq@Dev__F·
@bindureddy The "runs around in circles" thing is what gets me. If it can't even beat Opus on basic reasoning what's the point of all those extra compute cycles.
English
0
0
0
133
Bindu Reddy
Bindu Reddy@bindureddy·
Open 4.8 runs around in circles MORE THAN Opus 4.7 Seems like Anthropic loves to burn their GPUs cause they can 😂 TBH this was more of a press release than a model release Struggling to find a use-case where it makes sense to adopt it
English
49
10
261
17.7K
Umer Farooq
Umer Farooq@Dev__F·
@ajambrosino Searching actual chat content instead of just titles is the kind of update that makes you wonder why it wasn't there from the start. C
English
0
0
0
123
Umer Farooq
Umer Farooq@Dev__F·
@gdb Using the tool to build the tool. That recursive loop is where things get interesting.
English
0
0
2
1.2K
Umer Farooq
Umer Farooq@Dev__F·
@dkundel Codex handling thread orchestration sounds cool but I'd worry about losing visibility into what's happening under the hood.
English
0
0
0
82
Umer Farooq
Umer Farooq@Dev__F·
@simpsoka they look clean tbh, way better than generic avatars
English
0
0
1
78
Umer Farooq
Umer Farooq@Dev__F·
@timsoret GDPR really slows things down over there, it's a constant headache for rolling out new features.
English
0
0
0
35
Tim Soret
Tim Soret@timsoret·
Why the latest AI features (Computer Use, Advanced Voice, Agents, Memory, etc.) always land LATE in the EU… OpenAI, Anthropic, Google, Apple, Meta… they all do US-first (or select markets), then EU weeks/months later, often watered-down or geo-blocked at launch. 1. GDPR is the #1 killer
Any feature touching personal data hard (screenshots of your desktop, memory of past chats, voice biometrics, screen understanding, personalization) triggers massive privacy reviews, consent flows, data residency rules, impact assessments. Fines = up to 4% of global revenue. Easier to just flip the switch OFF for EEA users on day one. 2. EU AI Act is already biting
Frontier models + agentic tools (“computer use” that sees/clicks/types on your screen) = systemic risk checks, transparency obligations, high-risk classifications. Emotion recognition bans affected voice modes. The Act is phasing in and companies are being extra careful. 3. DMA + DSA pile on
Apple literally blamed the Digital Markets Act for delaying Apple Intelligence (“it would hurt privacy/security because of forced interoperability”). DSA adds extra safety/content moderation burdens for anything generative (deepfakes, illegal output, etc.). Real example right now:
OpenAI’s new “Computer Use” / desktop agent (the one that literally controls your macOS/Windows apps via screenshots + cursor) is explicitly unavailable in EEA, UK & Switzerland at launch. Same story with Claude’s advanced agents, Gemini features, Meta multimodal, etc. 
US rules = ship fast & fix later.
EU rules = legal sign-off first, then ship (sometimes nerfed). 
The market is too big to ignore forever, so features do arrive… just after the lawyers finish. Frustrating for those like me in Europe to watch everyone else get the life-changing, workflow-shattering new toys first.
English
4
0
13
3.7K
Tim Soret
Tim Soret@timsoret·
Meanwhile in Europe... We are literally forbidden by @EU_Commission & @GOVUK to use latest available technology everyone else enjoys, losing months of early adoption while the rest of the world is racing, making us less competitive.
Tim Soret tweet media
OpenAI@OpenAI

Windows users, this one’s for you. Computer use now works on Windows, so Codex can take action on your Windows computer. And with Windows support for Codex in the ChatGPT mobile app, you can start, review, and steer tasks on the go while work continues on your Windows machine. An early experience, but we’re working on more ways to keep your work moving, wherever you are.

English
11
0
44
10.5K
Umer Farooq
Umer Farooq@Dev__F·
@nickbaumann_ The CoS pattern is smart bc that one thread already has the full picture. Spinning up new threads from context it already gathered beats pasting the same background into every fresh session.
English
3
0
13
7.6K
Nick
Nick@nickbaumann_·
This has fundamentally changed how I use Codex - everything runs out of a single persistent thread (my "chief of staff") - anytime I start a new project or workstream, I have that thread spin up a new thread (because it's already found the context from slack, etc) - the CoS thread checks in on the project threads during heartbeats, and occasionally sends relevant updates from slack to that thread everything flows naturally to the top
Guinness Chen@guinnesschen

If you ever get tired of managing your Codex threads, just let Codex manage itself! Codex can now create threads, search them, organize them, pin the important ones, and spin up worktrees for parallel tasks.

English
95
113
2.5K
359.3K
Umer Farooq
Umer Farooq@Dev__F·
@timsoret @EU_Commission @GOVUK The regulatory lag is real, especially in AI. By the time something gets approved here, everyone else has already shipped and iterated twice.
English
0
0
0
54
Umer Farooq
Umer Farooq@Dev__F·
@simpsoka @fchalissery Ah the classic "why didn't this exist sooner" feature. This is the kind of thing users don't know they need until they need it, then they wonder how they lived without it.
English
0
0
1
48
Umer Farooq
Umer Farooq@Dev__F·
@OpenAIDevs stable identicons across tabs is one of those small fixes that saves so much context switching friction. reading agent names every time you switch gets old fast.
English
0
0
1
1.7K
OpenAI Developers
OpenAI Developers@OpenAIDevs·
2 quality-of-life improvements for developers working with Codex today: Codex background agents now have stable pixel identicons. When the same agent shows up across tabs, mentions, transcripts, and the thread panel, it’s easier to recognize at a glance.
English
114
96
2K
171.2K
Umer Farooq
Umer Farooq@Dev__F·
@elonmusk That single word carries more weight than most essays.
English
0
0
0
161

Descubrir

@gmi_cloud @bin_zeshan @daniel_mac8 @iruletheworldmo @danshipper @bindureddy @ajambrosino @gdb