Krish

1.5K posts

Krish banner
Krish

Krish

@krishv

Engineer passionate about tech shaping our future. AI & data enthusiast. Amateur photographer, avid traveler, always curious about the next big innovations.

United States Entrou em Nisan 2010
384 Seguindo225 Seguidores
Krish
Krish@krishv·
Observability tells you what the agent did. Governance tells you whether it should have been allowed. Production agents need identity, scoped permissions, policy enforcement, approval gates, and audit trails. Full breakdown: clype.io/blog/ai-agents…
English
1
0
0
15
Krish
Krish@krishv·
A support agent may read a customer record. Should it modify billing? A finance agent may draft a refund. Should it approve it? A coding agent may open a PR. Should it deploy to production? Agent identity and user identity are not the same thing.
English
1
0
0
6
Krish
Krish@krishv·
An AI agent with tool access is not a chatbot. It is a runtime with permissions. Once it can update records, query databases, trigger workflows, or call internal tools, the main risk is no longer just hallucination. It is authority.
Krish tweet media
English
2
0
0
15
Krish
Krish@krishv·
@JamesZmSun Why not Atlas as the driver? It is also a Chromium browser?
English
0
0
0
90
James Sun
James Sun@JamesZmSun·
Today, we are excited to introduce Codex for Chrome! Now, Codex can drive its own Chrome tabs in the background to automate tasks while you use the browser simultaneously. It does this by opening up tab groups for each task, cleaning up at the end, and handing back tabs for review only as needed. Try it for deep research inside logged-in websites, large scale data transfer into any systems of record like CRMs/CMSs, and automating repetitive workflows inside admin consoles & internal tools. Codex will still prefer dedicated plugins if you have them installed, but the Chrome plugin is the universal connector that glues end to end workflows where programmatic coverage is often incomplete. We are making this available on both Windows and Mac today! Let us know what you think.
OpenAI@OpenAI

Codex now works directly in Chrome on macOS and Windows. It’s even better at working with apps and sites in Chrome, and now works in parallel across tabs in the background without taking over your browser. To get started, install the Chrome plugin in the Codex app.

English
52
43
604
206.8K
Krish
Krish@krishv·
The AI race is becoming an infrastructure race. MRC is a good reminder that scaling frontier models isn’t just about bigger GPUs; it’s about keeping massive clusters synchronized with less wasted compute, better reliability, and open networking standards.
OpenAI@OpenAI

AI supercomputers need a new kind of network to stay in sync at massive scale. OpenAI’s @markjhandley and @poyntingatgreg join @AndrewMayne to discuss what it takes to move data across record numbers of chips reliably and efficiently, the new Multipath Reliable Connection (MRC) networking protocol, and why it's available for the whole industry to use.

English
0
0
0
20
Krish
Krish@krishv·
Production AI agents need a different observability model than traditional applications. Hard failures are not always API errors or latency spikes. They often happen earlier in the run: stale retrieval, wrong context, incorrect tool parameters, memory issues, or policy decisions that are hard to inspect after the fact. For agent systems, every production run should preserve prompt versions, retrieval lineage, tool inputs/outputs, policy decisions, evaluator signals, and downstream side-effect IDs. Not hidden chain-of-thought. Just the observable execution path required to explain what happened. Full post: clype.io/blog/productio…
English
1
0
2
38
Sam Altman
Sam Altman@sama·
GPT-5.5 is going to have a party for itself. it chose 5/5 at 5:55 pm for the date and time. if you'd like to come, let us know here: luma.com/5.5 codex will help the team pick people from the replies. 5.5 had some good ideas/requests for the party, which we'll do.
English
1.9K
378
6.2K
906.5K
Krish
Krish@krishv·
@gdb I now believe it after trying GPT-5.5, but why is it limited to my computer? Why can’t it be on the cloud, allowing me to break free from working on a desktop or laptop computer?
English
0
0
0
157
Greg Brockman
Greg Brockman@gdb·
this has one of the most exciting launch weeks in OpenAI's history, with a goal of making agents more real, useful, and accessible for all our users. codex can now smartly do much more on your computer, remember more of your context, and run more ongoing work independently.
English
180
117
3.1K
165.1K
Krish
Krish@krishv·
With agents transforming software product development, building is now easier than getting it right. Even if right, frontier orgs and enterprises take over the market fast. The hardest part is to disrupt without being swallowed by big tech.
English
0
0
0
29
Krish
Krish@krishv·
@budapp Interesting, will give a try but first I need to list down my use cases.
English
0
0
0
4
Bud
Bud@budapp·
Introducing Bud. The first AI Human Emulator. Bud has a full computer with storage, compute, and memory to build and code, sms and telegram to communicate, a full browser to use, can create/store/edit files, connect and use your tools, learn custom skills, work fully autonomously, and complete any task end to end just like a human. Text the number below or try free at bud [dot] app. Comment for 100k free credits.
English
2.8K
323
4K
698K
Krish
Krish@krishv·
I asked different models to solve a puzzle with the same exact inputs. It's interesting to see Grok 4.20 one-shot it, while Claude Opus 4.6, Gemini 3.1 Pro, and GPT-5.4 Thinking couldn’t get it until I manually drove them to the solution with prompts. I'm impressed by Grok's critical thinking ability.
English
0
0
0
56
Krish
Krish@krishv·
@elonmusk This is amazing, but what happens to the HW3 equipped Teslas?
English
0
0
2
31
Elon Musk
Elon Musk@elonmusk·
Tesla V14.3 self-driving review. The point releases will bring polish. V15 will far exceed human levels of safety, even in completely unsupervised and complex situations.
Zack@BLKMDL3

600 miles in with FSD v14.3 already and here are my thoughts: The improved reaction time is immediately noticeable and definitely quicker than a human could react. Yesterday a semi truck swerved fast into my lane and the car reacted insanely quickly to get me around them. Tesla says it’s 20% quicker but feels more than that- and it was already extremely good. I’ll start by saying FSD was already so great with v14.2 that it’s sometimes hard to find new things, but there some huge apparent changes immediately noticeable with v14.3. To those who think it’s not a big improvement over v14.2.x+, you’ll be very impressed and especially with some more polish. The reinforcement learning upgrades and thinking are noticeable. Parking is where you immediately notice some changes. In the release notes it says that parking is quicker and more decisive and it’s true. It has picked spots closer and quicker to the selected pin. My Y isn’t showing the new P pin graphic for parking pin for some reason, but it’s definitely parking closer than before with more thought to it. Looking forward to eventually getting more options to hopefully park either closer or further away from people. For the very first v14.3+ build, I have to say it’s pretty polished. The only gripe I have is the way it won’t get out of the left lane soon enough on highways. It likes to cruise in the left lane which isn’t ideal, it’s gotten better but the addition of reduction in unnecessary lane changes needs to be dialed back a bit. Lane changes are a huge plus with this build and they are quick, decisive and executed very well, smooth as butter too. Turn signals come on at way better times now in parking lots and at the perfect time on the road. I was lucky enough to get the update with 600 miles left of my 1800 mile Oregon road trip, so I pulled over to install it so I could get as much experience as possible with it to share with you all. So far 600+ miles in, I’m impressed. A few rough edges with the left lane behavior and the last few inches of parking are a bit slow 1/5 times until it puts it into park but with a point release update everything should be dialed in. The 350 mile drive home today from the Bay Area had zero intervention including all parking and charging. One thing I would love to see implemented is a reset button for the FSD stats page. Would be cool to have a specific trip meter for FSD stats on Trip A/B or make your own trip. I’ve been hinting at a pretty cool road trip next month with my 2025 Model 3 so it would be cool to have a reset for that. Speed control seems good on highways, it’s matching traffic speed great. Braking is very impressive for sudden slowdowns, had a big one in San Jose last night it did a great job with from 80-10mph. HUGE improvements with stop sign behavior. The acceleration and deceleration are way smoother than before, much more pleasant. Mad Max takes off strong but again, a better curve than before. Mad Max is also polished a bit and feels great. FSD v14.3 did a great job in LA traffic once I got back and will go out this evening to film videos for everyone on my normal test loops. Let me know if there’s anything specific I should try or check out. Can’t wait to see how v14.3+ progresses especially with the upgraded reasoning coming to all scenarios soon. Some awesome additions here. THANK YOU everyone @Tesla_AI for all the hard work getting this update out. More videos to come.

English
2.3K
4.2K
24.4K
8.5M
Krish
Krish@krishv·
@elonmusk @ai_for_success I believe Grok and xAI are underrated. Using them for information with real-time user sentiment is way better than others.
English
1
0
0
61
AshutoshShrivastava
AshutoshShrivastava@ai_for_success·
Google DeepMind is destined to win the AI race.
English
177
74
2.2K
585.3K
Krish
Krish@krishv·
@garrytan I believe there is a fundamental misunderstanding about the concept of instructing coding agents. This represents a new form of communication with programs that require guidance to complete their tasks. Consider it like working with interns just beginning their journey.
English
0
0
1
71
Garry Tan
Garry Tan@garrytan·
The thing I believe that few people believe but I think everyone will believe Markdown *is* code
Garry Tan tweet media
English
228
59
816
167.4K
Krish
Krish@krishv·
The way we interact with these agents will be transformative, yet it will largely revert to traditional usage patterns due to established UI/UX conventions. The two unifying factors are tracking and, above all, voice as the primary input method.
Andrej Karpathy@karpathy

@nummanali tmux grids are awesome, but i feel a need to have a proper "agent command center" IDE for teams of them, which I could maximize per monitor. E.g. I want to see/hide toggle them, see if any are idle, pop open related tools (e.g. terminal), stats (usage), etc.

English
0
0
0
28