AI/ML • VLSI • RISC-V Building things, breaking things, figuring it out

Earth
Joined November 2019
rohan retweeted
MiMo-V2.6-Flash officially supports a 1M-token context window, but OpenCode currently shows 200K. Is this an OpenCode limit or a model metadata mismatch? 🤔
1
1
22
MiMo-V2.6-Flash officially supports a 1M-token context window, but OpenCode currently shows 200K. Is this an OpenCode limit or a model metadata mismatch? 🤔
1
1
22
rohan retweeted
Replying to @0xSero
yep, they’re peak
1
2
670
rohan retweeted
using AI to research how to make money online using AI 😭
1
1
17
rohan retweeted
Finally bought a brand new 16-inch MacBook Pro with the M5 Max chip What should I run first? 😭
1
1
1
66
Finally bought a brand new 16-inch MacBook Pro with the M5 Max chip What should I run first? 😭
1
1
1
66
using AI to research how to make money online using AI 😭
1
1
17
me: I’ll just learn AI also me 3 weeks later: reading a 40-page accelerator paper at 2 AM wondering why the psum buffer exists 😭
1
22
bro,what the hell glm 5.3 flash is even doing?
1
30
Thanks to the 600+ replies from everyone on what we can do to improve the Gemini API + AI Studio Platform. Here’s the Top 9 stack rank from your feedback so far: 1) Free Tier Rate Limits & Credits: This was the clear #1 request. You want to be able to use our Flash and Pro models in agentic flows on our Free Tier. Moving away from RPD to Credits. 2) Use your subscription on the API: Have a paid token allowance from your Google AI Pro and Ultra subscriptions on the API. 3) Real-Time Usage API: Programmatic visibility into token burn, spend, and rate limits. Set spend limits in request and get spend per request in response body. 4) Billing is confusing: Charts not updated in realtime, and users are forced to go into GCP Billing console for some flows. Also Prepaying isn't clear. 5) Gemma TPM: Agent harnesses are hitting 16k TPM far too quickly. You want to build more with Gemma on Gemini API. 6) Richer API errors: Failed tool names, arguments, raw response, trace IDs, token offsets, etc. Quota-aware retries. 7) Content moderation: Media models are being aggressive with blocking queries. 8) Reliability: Spikes of 503s for lower Tier users. 9) Tool loading: Allow progressive tool loading (adding new tools without busting cache). Keep it coming!
138
24
8
796
69,186
Just ran out of OpenAI credits 😭 Need a solid Codex replacement for the next 5 days. What’s everyone using right now that’s actually good for coding?
53
rohan retweeted
How to Get 17M+ Views on X In 4 Days ? Be @moonpay > Launch your first airdrop and give users an average of $14 for free > Crypto creators start posting, and you get free marketing > Announce a second airdrop and keep posting > Suddenly, every tweet starts getting millions of impressions. Farm impressions all day and pull 17M+ across 6-7 posts > Then announce the second airdrop with an average reward of just $5–10 > Introduce mandatory KYC and ban 90% of the countries whose users claimed the airdrop > Money saved > Views and engagement farmed > marketing achieved A genius at work 😭
36
3
85
6,401
rohan retweeted
U are very clever intern
over 400 people have claimed so far and over $4,000 has been delivered we still have over $220,000 to give out there are a ton of people trying to participate and the team is actively working on improving the experience thank you for your patience
111
2
168
8,501
USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH USE 5.6 SOL ON HIGH
1
62
Replying to @thsottiaux
Coming soon? Baby OpenThropic ClauDex?
13
13
7
608
36,277
rohan retweeted
🚨 A severe issue is currently happening with Codex. The Codex app on Mac is overloading, causing permanent damage to SSDs and overall device lifespan. Multiple users have already reported that their SSDs completely fried, forcing them to get repairs. This issue was raised weeks ago, and OpenAI claimed they fixed it. Yet, a massive number of users are still reporting the exact same problem right now. Source: Reddit ⬇️
326
201
125
2,057
555,036
rohan retweeted
India’s IT ministry banned Telegram for one week because some users shared leaked exam questions. This punishes 150M+ ordinary Telegram users in India — not the insiders who leaked the exam materials. And the ban hasn't stopped anything. The leaks just moved to other apps.
Statement : Shutting down Telegram is a band aid solution and is a disproportionate answer to exam fraud The Internet Freedom Foundation objects to the directions announced today in the National Testing Agency's press release on action against the Telegram platform. On the NTA's recommendation, the Ministry of Electronics and Information Technology has, under Section 69A of the Information Technology Act, 2000, restricted access to the whole of Telegram in India until 22 June 2026, and has separately ordered the platform to switch off message-editing for every Indian user until 30 June 2026. This is a blunt, nationwide measure aimed at the conduct of rampant fraud rackets, and on the Government's own admission is constitutionally incompatible. At the outset it is important to note that Section 69A and the Blocking Rules of 2009 framed under it allow the Government to block access to specific “information” on a computer resource. They do not extend to switching off an entire intermediary, still less to ordering a company to redesign its product by removing a feature for a whole country. In Shreya Singhal v Union of India, the Supreme Court upheld Section 69A because it is narrow and hedged with procedural safeguards. Reading it to authorise shutting down a platform that lakhs use is an overbroad restriction by the NTAs own admission. For the message-editing direction the release identifies no source of power at all. If one exists, the order must say so. The release argues against itself A restriction on access has to be the least intrusive measure that achieves its aim as per the constitutional test of proportionality laid down in Justice K.S. Puttaswamy v. Union of India (2017) and applied in Anuradha Bhasin v. Union of India (2020). The NTA's own narration shows the block fails its nodal agency, the release says, “has secured the prompt take-down of a substantial number of Telegram channels, groups and bots”, and this targeted work “is the reason the harm caused by these rackets has been contained to the extent it has”. If channel level takedown contained the harm, the case for a blanket block collapses and hence the Government has reached for a heavier tool while conceding that a lighter one was working. The collateral cost sits on the record too as noted in the press release. The block, the NTA accepts, “affects lakhs of citizens who use the Telegram platform for legitimate personal, educational, professional and informational purposes”. The release also says there is "no such paper available outside the secured examination chain" and that “the security of the examination is unaffected by the action taken”. If the exam is secure and no leak exists, what is being suppressed is rumour, and rumour cannot justify closing a platform when specific blocking and criminal prosecution remain available. Students use of Telegram The block of telegram is reactive and ineffective and will punish ordinary users instead of addressing the systemic source of exam leaks. This blocking comes in the final days of NEET preparation, when thousands of students depend on Telegram for study groups, doubt-clearing, and shared resources. Also, it is important to consider that the source of exam papers leak will occur from inside the system, among insiders and across the printing and logistics chain, with the platform being the most downstream channel for distribution. Hence, switching off Telegram, is merely a deflection from the repeated failures that will continue while media attention is directed towards this Telegram ban. Lack of transparency At present only a press release from the NTA has been provided, which recommended the block but the reasoned order of MeitY, the authority that issued it, has not been released. The Anuradha Bhasin decision requires that orders restricting access be published so they can be tested in court. Here the order, and the reasoning of the committee behind it, stay out of view, and we do not know whether Telegram was heard at all. An announcement of a block is no substitute for an order the affected party can challenge. Blunt to enforce and very easy to evade Usually, app-level blocks run through IS-level DNS and IP filtering. They are over inclusive, sweeping in lawful use, yet simple to evade as a determined exam leak racket moves to a VPN or a mirror within minutes while ordinary users lose the service for a week. We ask the Government to: 1) Publish the MeitY Section 69A order and the NTA recommendation behind it, with reasons; 2) State the legal basis for the message editing direction, or withdraw it; 3) Confirm whether Telegram was given a hearing under the Blocking Rules, and place the committee's record before any court that hears a challenge; and 4) Lift the platform-wide restriction and rely on the targeted takedowns the NTA itself credits with containing the harm. We emphasise that the NEET (UG) 2026 re-examination is worth protecting and it concerns the future of lakhs of aspirants. It requires securing the entire process of examination rather than reaching for purported band aid solutions that instead cause more harm. The State cannot switch off a service used by lakhs to answer the wrongdoing of a few, and cannot do it through an order no one affected is allowed to read. On its own facts, the Government has done both. New Delhi, 16 June 2026.
3,414
11,511
2,000
55,361
7,021,591
That why I hate reservation My family income - under 12 lakh My college fees - 1 lakh My PG fees - 70000 My sister's college fees - 1 lakh Reservation=NO My ST community classmate His family's income - 60 lakh His college fees - 5,500₹ His hostel fees - 1000₹ His sister's college fees- 5,500 His sister's hostel fees - 1000₹ Reservation- super YES ! #endreservation
70
105
11
616
55,900
Kimi K2.7 Code just shipped. Open source with stronger coding, better agent performance, and ~30% fewer reasoning tokens than K2.6. Improvements: • +21.8% Kimi Code Bench v2 • +11% Program-Bench • +31.5% MLS Bench Lite Open-source models keep raising the bar. 🔥
90
rohan retweeted
Fixed it for you
1
3
243