Thriving upon the curiosity with(in) myself
Joined January 2023
- Tweets1.1K
- Following238
- Followers33
- Likes1.1K
Pinned Tweet
Personal Update:
>2nd year, 3rd sem
>Primarily Doing) DSA, OOPs in CPP
>In process of finding a team of like minded people to grind with
>Currently going through SIH problem statements
>Will be participating in multiple hackathons this year
>Hmm what else?Ig Just wish me luck
Sure can just leave a seperate browser for the agent
OpenCode Browse is now available
it lets your agent control your real browser with your tabs and logins and extensions
opencode.ai/product/browse
It has been improving continuously ever since the rebranding and now we get this:
𝕏 just rebuilt its entire web app from scratch, and the performance improvements are absolutely insane
On a slow connection, the old 𝕏 web app took 23.9 seconds to display the first post. The new version does it in just 2.7 SECONDS
That's nearly 9 TIMES faster
Here's what changed:
• Three separate web-serving systems, serving users, search crawlers and logged-out visitors, have been consolidated into ONE unified app
• On a slow connection, a SpaceX post loads 12.8x faster and the SpaceX profile loads 8.2x faster for signed-in users
• Profiles and posts now appear before JavaScript finishes downloading, because the content is already included in the HTML
• Just FIVE engineers working alongside SI coding agents built the first version in under six months
• Logged-out 𝕏 and search crawlers are already running on the new architecture, while signed-in 𝕏 is being rolled out country by country
And the scale of this achievement is extraordinary
𝕏 is the sixth most visited website in the world, serving an enormous global audience across different devices and network conditions
This rebuild is bringing dramatically faster loading times, especially for people using slower internet connections
And what's even more impressive is that such a small engineering team accomplished this with the help of SI coding agents
blog.x.com/en/engineering/re…
calls the pendant hunt a dead end then teases a thing with no screen that couldn't be an app 👀 it's a pendant isn't it?
Let's see what you've cooked this time around 👀
when bend2 launched, some complained about AI usage. in my view, this is only a problem if bugs can occur. the current typescript compiler, for example, has many. I hate it. with well-designed laws, though, that ceases to be a problem. so, for the better or for the worse, bend3 will be 100% AI-written. there is no going back
while I was raising, I left a fleet of agents working on bend3, a small scale version of this vision. this was a huge success, beyond my own expectations. opus 5.5 successfully rewrote all of bend in itself, including its compiler, which is now certified, with zero bugs, and already 2x faster (!!). we now have a model of Bend, C, CUDA and the compiler, all written in Bend itself, and properly formalized. with this, we'll soon start the largest proof-driven compiler self-improvement loop in history
news on Bend have been slow because I was busy raising, but as soon as this is over, expect updates at a breath-taking place. this is the most exciting thing I've ever done and I can't wait for the big run to start
Will improve it further
A friend of mine has built something genuinely interesting!! 👀
Meet Bell AI - an AI-powered voice assistant designed to help elderly people live more independently.
It makes automated calls to remind them about medicines, medical appointments, and other important tasks, while allowing them to interact through natural voice conversations and confirm task completion.
The idea is to make everyday assistance more accessible without requiring elderly people to navigate complicated apps or interfaces.
Built by @_karsh0506 @itzdeveshvaidya and @ayeayushh ❤️
Check it out: bellai-app.vercel.app/
Do give it a try and share your thoughts with the team!! 🤝
Incredible
$3.46 TRILLION has been wiped out from AI stocks in October.
1. Intel: -43.4%
2. Coreweave: -37.7%
3. Nebius: -36.7%
4. SanDisk: -33.4%
5. Seagate: -30%
6. Marvell: -25.8%
7. Dell: -22.3%
8. SKHY: -19.8%
9. WDC: -19.2%
10. Google: -18.7%
11. Micron: -17.8%
12. AMD: -17.2%
13. Meta: -15.8%
14. TSMC: -11%
15. NVDA: -10.7%
16. SpaceX: -9.4%
17. Broadcom: -8.6%
18. Oracle: -4.3%
nvidia's agents are steering a 42 dish telescope looking for aliens and mine [ate the last 30% of my usage looping on a failed cloud run] 😭 xitter.gq/NVIDIAAI/status/211656…
AI agents are now steering the Allen Telescope Array in Northern California to search for possible alien signals. 👽
Happy Halloween. Sleep tight. 🎃
See how the @SETIInstitute and Breakthrough Listen are using NVIDIA Holoscan and CUDA-X libraries to listen to the skies → nvda.ws/4zgYoGv
That's somewhat the overview of this. Tho the downside is that the existing subscribers are suffering cuz of this, which isn't ideal at all
𝗧𝗵𝗲 𝗲𝗰𝗼𝗻𝗼𝗺𝗶𝗰𝘀 𝗼𝗳 𝗚𝗿𝗼𝗸 𝗕𝗼𝘁 𝗮𝗿𝗲 𝗯𝗲𝗶𝗻𝗴 𝗺𝗶𝘀𝘂𝗻𝗱𝗲𝗿𝘀𝘁𝗼𝗼𝗱
At $200/month and 800K paying subscribers, Grok Bot generates $160M/month, or $1.92B annualized.
The problem isn't necessarily revenue. It's the cost of serving those subscribers.
Grok Bot is reportedly routing requests to Anthropic's models, including Opus 5.5.
Think about the economics.
SpaceX is reportedly renting compute to Anthropic at $50B/GW annually, or $50M/MW.
Meanwhile, SemiAnalysis estimates Anthropic generates $102.9M/MW/year in API revenue at 92.2% gross margins.
Anthropic subscriptions generate just $16M/MW/year at 50% gross margins, despite consuming 41.7% of inference compute.
The same compute generates substantially more revenue through API customers than through flat-rate subscriptions.
Now consider Grok Bot.
If you're charging $200/month for unlimited or heavily subsidized access while serving expensive frontier models at API-equivalent inference costs, usage can quickly consume the subscription revenue.
You're effectively buying expensive inference at retail prices and reselling it through a fixed-price subscription.
And that compute has an alternative use.
SpaceX can rent it to Anthropic, generate attractive infrastructure revenue, and use the proceeds to train Grok.
Anthropic's subscriptions make strategic sense because they drive distribution and generate valuable training data. But they're still leaving substantial money on the table relative to API pricing.
The logical strategy for SpaceX is to monetize compute externally while Grok catches up.
Once Grok 5 reaches competitive inference economics, bring that capacity back.
The goal isn't to maximize Grok Bot usage today.
It's to maximize the return on every watt until Grok can generate a higher return than renting the same compute to Anthropic.
That's the advantage of vertical integration.
Yeah as long as you work in company that has a frontier model and they give you unlimited usage
Weekend hack: HackerTube. I was annoyed with YouTube's recommendations so I had Claude filter down 43M comments over 20 years of HN to 6,454 videos that keep being recommended. *Significantly* better viewing experience, ~2 hours with Opus 5.5:
hackertube.org/
- Excellent recs without any logins (state is kept in local browser session)
- Thumbnail is a representative frame from middle of the video instead of the super duper A/B test hyper-optimized user uploads
- Underlying data from the comment analysis as one zip file, point your Claude at it and ask for your own version
- HackerTube Longs ✊
A bit more on the process:
I use Claude Code (desktop app for Mac), Opus 5.5 on Medium effort as my default. I only talk to Claude and have it do everything, the only "manual work" I helped with in tandem was in the browser (reminiscent of MenuGen): 1) I had to create some kind of API key from Google Cloud UI to query YouTube for video metadata, 2) I booted up a VPS box for serving and 3) I bought the domain name. Once the box was up I gave Claude the IP and told it update, secure, harden, deploy and optimize the code, and it did the whole thing.
Many awesome projects that would have normally taken weeks/months even can now be few-shotted for breakfast. What else are we not thinking of.
wait on opus 4.6 and sonnet 4.6 it only applies at default temperature, top_p and top_k??
so nudge one setting and the watermark's gone 💀
Anthropic adds its invisible watermark to 4 more Claude models on Tuesday.
Its customer email names Mythos 5, Opus 4.7, Opus 4.6 and Sonnet 4.6. Opus 5.5 already has it.
→ Claude hides the pattern in its word choice and adds no characters
→ Anthropic says nobody can trace it to you, your company or your chat
→ Editing a few lines probably won't remove it
→ Approved regulators, media and researchers can check for it
Would you tell a client if Claude wrote their copy?
lol my bots were looping on the cloud while I was busy with something else and then when I came back found that they drank up the last 30% of my remaining usage, all while just making a couple of changes in an existing website
And the reason I am highlighting this? cuz I am on super tier. The last few days I have done so much work which barely affected the usage status. Nevertheless the limit resets tomorrow after noon so until then lemme see what caused this...
See if you find something useful here @poteto
Burned through 800.7M tokens in like 6 days, with 44.5%(356.3M tokens) wiped out in just 12 hours today
(BTW I am on super)
Sharing some telemetry from my usage logs vs. Cloud Agents runs that might help the team optimize agent retry loops and context window compaction:
> Oct 25–29 (Local Grok Bot + Cursor): ~247.5M tokens (0 Cloud Agent runs on Oct 28–29)
> Oct 30 (10 Cloud Agent runs, 60% success): 196.8M tokens
> Oct 31 (2 Cloud Agent runs, 12 hrs): 356.3M tokens
Breakdown of where the quota actually went 👇
