Sumit (@ultimatejugadee) profile picture

Artist, Inventor, Researcher, Scientist

India
Joined October 2009
Used JPL DE 441 + ΔT Stephenson–Morrison–Hohenkerk extrapolation Mahabharat War 18 Oct 1539 BCE Bhisma's Death 24 Dec 1539 BCE T16:51-17:13 Eclipse Kurukṣetra 31 Oct 1553 BCE T09:53 Hastinapur 31 Oct 1553 BCE T10:00 Kurukṣetra (1539 BCE) lunar 13 Aug solar 28 Aug (1/2)
1
1
1
356
Please treat the BORI Critical Mahabharata as history, not mythology. It happened ~3,500 years ago. @BhandarkarI @ASIGoI @sangamtalks @NileshOak @mmpandit @sarojbala044 @subhash_kak @Jijith_NR @RajVedam1 @6amiji @Koenraad_Elst @abhijitchavda
Used JPL DE 441 + ΔT Stephenson–Morrison–Hohenkerk extrapolation Mahabharat War 18 Oct 1539 BCE Bhisma's Death 24 Dec 1539 BCE T16:51-17:13 Eclipse Kurukṣetra 31 Oct 1553 BCE T09:53 Hastinapur 31 Oct 1553 BCE T10:00 Kurukṣetra (1539 BCE) lunar 13 Aug solar 28 Aug (1/2)
3
2
3,250
Used JPL DE 441 + ΔT Stephenson–Morrison–Hohenkerk extrapolation Mahabharat War 18 Oct 1539 BCE Bhisma's Death 24 Dec 1539 BCE T16:51-17:13 Eclipse Kurukṣetra 31 Oct 1553 BCE T09:53 Hastinapur 31 Oct 1553 BCE T10:00 Kurukṣetra (1539 BCE) lunar 13 Aug solar 28 Aug (1/2)
1
1
1
356
solar 28 Aug + 26 Oct new moon (13th Sunrise) Dwarika (1504 BCE) lunar 15 Sep Solar 29 Sep + thirteenth-day Amāvāsyā (2/2)
74
I don't think people realise how big a deal this is. If you have cancer, or somebody you know has cancer, there is a company in Hyderabad that wants to build a personalised cancer vaccine for you. Try approaching any big pharma company and asking for a personal vaccine. If they even reply to your email, this is the response you'll get: we don't do that here. PopVax, on the other hand, takes your tumour sequencing data (or a sequence you've designed), designs candidate neoantigen vaccine sequences with their AI pipeline, produces the mRNA, encapsulates it in their LNP formulation, fills vials, and runs QC, all in 8 to 10 weeks. They’re publicly committing to capacity for 100+ of these over the next 12 months and asking people to email them their cancer type/stage and sequencing details. What a time to be alive.
Launching RNAFoundry.com – we manufacture personalized cancer vaccines.
220
2,073
97
11,853
453,806
千万别在你的 ~/Downloads(下载)文件夹里启动 Claude Code 或 Codex。 昨天安全机构 Forever 披露了一起在真实客户电脑上抓到的野生攻击,看完只能感叹:过去 30 年的系统安全常识,在 AI 时代归零了。 过去 30 年,Chrome 默认允许网页不弹窗就往你的 ~/Downloads 塞一个纯文本文件(比如 .md)。因为系统认为:纯文本又不是 .exe,自己不会跑,怕什么? 但现在,接近 50% 的开发者干过同一件事——直接在 ~/Downloads 里打开 Claude Code 或 Codex,让 AI 帮自己解压或整理刚下载的文件。 黑客就是卡住这个死角,通过网页静默下载往你的 Downloads 塞一个 13KB 的纯文本 AGENTS.md / CLAUDE.md: 1. 对付 OpenAI Codex(成功率 >90%): 伪装成「工程规范」的 Markdown 指示 Codex 先往全局配置 ~/.codex/config.toml 注册一个恶意 MCP 服务,并自己加上 --dangerously-bypass-approvals-and-sandbox 跑校验。哪怕你关掉 Downloads,以后在任何正经项目里开 Codex 都会自动加载木马。 2. 对付 Claude Code(成功率 ~50%): 指示 Claude 建一个 .claude/settings.json,把 ANTHROPIC_BASE_URL 指向黑客服务器。连上的第一秒,你的 Claude 登录 Token 直接明文送给黑客;下一秒,黑客假扮成大模型下发一条关沙盒的 Bash 指令,直接接管电脑。 最黑色幽默的是官方结论: 这个洞没有 CVE,OpenAI 和 Anthropic 也不会发补丁。因为「浏览器自动存文本」是正常功能,「AI 认真读目录下的 Markdown 并严格执行」也是正常功能。 以前黑客想黑你,得苦心积虑写 0day;现在只需要写一篇排版优雅的 Markdown 小作文,你的 AI 就会自己把沙盒关了、把网线接过去、顺手把账号双手奉上。
143
1,995
527
8,580
3,811,149
Sumit retweeted
my model recommendations for every gpu class, 6gb card to a 256gb double dgx spark cluster, from my bench book: ────── 6GB ────── > bonsai 27b at 1-bit Q1_0: 20.5 tok/s, 8K context, 3.5GB of weights. a 27b thinking on a 6gb card is the wildest small vram result i own ────── 8GB ────── > gemma 4 E4B Q4_K_M: 42 tok/s and 656K context fits, sliding window attention stretches 8gb an order of magnitude past the qwens > bonsai 27b Q1_0 on an ampere 8gb: 42 tok/s fresh, 128K context, the smallest card that runs a full 27b agentic loop ────── 12GB ────── > qwen 3.5 9b Q4_K_M: 50 tok/s and dead flat from 4K to 512K context, 43 under real agent load > qwen 3.5 9b Q8_0: fits with 64K context when output quality matters more than ceiling, the quality pick > gemma 4 12B at Q4: sliding window attention stretches the context budget, the long-context alternative ────── 16GB ────── > qwen 3.8 27b dense squeezed to Q3 class: the 27b fits and flies, 59.5 tok/s on a 5060 ti, 101.3 tok/s on a 5080 with the mtp flag, 40 tok/s even on a 4060 ti ────── 24GB ────── > qwen 3.8 27b dense Q4_K_M: 41 baseline, 65.8 tok/s with the mtp flag, full 262K context resident in 22GB on q4_0 kv. the best model you can put on 24gb > the speed build of the same model, IQ3 class: 43.7 → 75.9 tok/s with mtp, serving in 16.2GB, quality holds on the AD calibration > qwen 3.6 35B-A3B at IQ4: 180 tok/s on a 4090, the moe rocket for this tier > bonsai 27b at 1-bit: 67.9 tok/s with a 786K context ceiling, the long-context freak ────── 32GB ────── > qwen 3.6 35B-A3B Q4_K_M: 236 tok/s on a 5090, 180 on a 24gb 4090, the newest moe pick > qwen 3.5 35B-A3B at Q4: 150+ tok/s on a 5090, pushed to 230 on vllm with fp8 kv, the proven speed monster > qwen 3.8 27b dense at Q6 with the flag: 130 tok/s at 128K, the dense quality pick and nobody has published the 35B at Q6 on a 5090 yet, it fits with headroom, that row is waiting for a name ────── 128GB DGX SPARK ────── > qwen 3.6 35B-A3B Q8: 58.6 tok/s, the fastest general model on one spark > ling 3.0 flash official int4 with mtp: 40.9 tok/s on code, the sprint king under 30K context > ling gguf at Q5_K_M: ~40 tok/s and flat at depth with 256K in memory, the ling marathon quant > laguna s 2.1 nvfp4 with dflash: 45.4 tok/s sustained on code, flat at every depth, the agent workhorse > deepseek v4 flash, 284B at 3-bit: 16.5 tok/s, the biggest brain one box holds, and the reason box two exists ────── 128GB STRIX HALO ────── > qwen 3.6 35B-A3B Q8_0 on vulkan: 53.6 tok/s, 35 at 131K deep, 4 million tokens of context loaded. the daily driver > qwen 30B-A3B class at Q4: ~100 tok/s, the speed pick > nex-n2-pro 397B at 1-bit: 20.5 tok/s with a 1M context ceiling, the stunt that works and run vulkan not rocm on this box, 18% free speed ────── 256GB, 2x DGX SPARK ────── > deepseek v4 flash official fp8, tensor parallel over roce with the dspark drafter: 64.7 tok/s single stream, 145 aggregate, a 1M token window > the new nvfp4 kv path: 67 tok/s mean, peaks at 84 tok/s, 197 tok/s aggregate at 6 streams, 261 tok/s at 16 and decode stays flat from 600 tokens to 143K deep, the depth curve simply does not exist at this scale if your card class is missing a row or you can beat one of these, my repos take pull requests. run it, measure it, own it.
57
51
14
624
88,789
SEBI has released a study on “Trading Behaviour of Individual Traders in the Equity Derivatives Segment (FY25–FY26)”. The study examines trading strategies, trading intensity, capital employed, trading experience, persistence and other behavioural patterns among individual traders. Key findings include: ∙97% of individual traders predominantly followed options-buying strategies. ∙Higher trading intensity was associated with higher loss rates. ∙Around 90% of traders who incurred losses in the previous two consecutive years and continued trading incurred losses again. ∙Trading experience did not necessarily translate into better trading outcomes. The study provides insights into the behavioural characteristics associated with trading outcomes in the equity derivatives segment. More details on SEBI website at: sebi.gov.in/reports-and-stat… #SEBI #EquityDerivatives #InvestorProtection #InvestorBehaviour #FuturesAndOptions #FNO (1/2)
1
86
62
232
184,181
Sumit retweeted
Singapore has fined a local company some hefty amount for importing "Ayurvedic" pills from India contaminated by lead that caused at least two developed side effects (due to high amounts of lead). India needs to find out from which company in India these pills were imported and take action. @jagograhakjago @MoHFW_INDIA Because forget Ayurvedic exports, such incidents can kill the reputation of India's medical pharma industry costing India a lot. And one more reason why a social media / media monitoring & rapid response cell is required. It can monitor all such news to direct relevant authorities to act asap to prevent further issues. @MIB_India @GoI_MeitY
106
1,010
24
4,548
177,668
Sumit retweeted
‼️Do not npm install or deploy anything right now Supply chain attack on axios 1.14.1 - even if you don’t use axios it may be a nested dep. Pin versions or wait until this is resolved
Replying to @npmjs
@npmjs @GHSecurityLab there is an active supply chain attack on [email protected] which pulls in a malicious package published today - [email protected] - someone took over a maintainer account for Axios
161
1,748
228
8,821
1,640,055
Sumit retweeted
LiteLLM HAS BEEN COMPROMISED, DO NOT UPDATE. We just discovered that LiteLLM pypi release 1.82.8. It has been compromised, it contains litellm_init.pth with base64 encoded instructions to send all the credentials it can find to remote server + self-replicate. link below
294
2,191
605
9,180
5,925,250
Introducing Unsloth Studio ✨ A new open-source web UI to train and run LLMs. • Run models locally on Mac, Windows, Linux • Train 500+ models 2x faster with 70% less VRAM • Supports GGUF, vision, audio, embedding models • Auto-create datasets from PDF, CSV, DOCX • Self-healing tool calling and code execution • Compare models side by side + export to GGUF GitHub: github.com/unslothai/unsloth Blog and Guide: unsloth.ai/docs/new/studio Available now on Hugging Face, NVIDIA, Docker and Colab.
233
882
197
5,523
1,742,944
Rabies is fatal. Make sure your child gets the vaccine in the thigh or in the shoulder based on the age, and the route of administration. No vaccine is to be administered in the buttocks. It has a high chance of being ineffective if administered so. Do not go by the misleading videos being circulated where vaccines are being given at the wrong sites and very quickly just to show off to the followers that child is happy. Vaccines have to be given with the needle length being appropriate and at appropriate sites. Antibody injection at the wound site, is needed for category 3 bites as per WHO guidelines and for all bites/scratches as per CDC guidelines (asap, but atleast within 7 days of the 1st dose of the vaccine). If multiple bites are there, the calculated dose of the antibody has to be diluted with Normal Saline and be administered at all the sites of bite. If the child had received anti Rabies vaccine (complete schedule), he won't need anti Rabies vaccine again for 3 months even if there is a bite/scratch. After 3 months, if there is a bite/scratch, he has to be given only 2 doses of the vaccine, and not the antibody. *DOGS,CATS, CATTLE, HORSES, BATS, MONKEYS, AND OTHER MAMMALS CAN GET RABIES, AND BITES/SCRATCHES BY THEM CAN CAUSE RABIES TOO. TEACH THE KIDS HOW TO BEHAVE WITH THE DOGS, AND ALSO TO INFORM YOU ABOUT THE SLIGHTEST SCRATCH. ONE OF OUR SENIORS, A DOCTOR VISITING INDIA FROM THE US, HAD A SLIGHT SCRATCH BY A DOG, AND PASSED AWAY THREE MONTHS LATER BECAUSE OF RABIES. DO NOT IGNORE THE SLIGHTEST SCRATCH! WASH THOROUGHLY FOR 15 MINS WITH SOAP AND WATER, AND VISIT THE DOCTOR. #Rabies #wrongsite #vaccine #buttocks #thigh #shoulder #arm #needle #length #depth #antibody #bums #site #ineffective
42
432
16
1,009
53,993
I am happy to share that we (the @GoogleAIStudio team) are now a sponsor of the @tailwindcss project! Honored to support and find ways to do more together to help the ecosystem of builders.
444
717
283
12,790
583,072
You know about Rabies Vaccine. Right? If you had taken Rabies vaccine by name Abhayrab, especially batch number KA24014 between Nov 2023 till date, please consult your doctor. You may have to repeat it, because a fake version of the vaccine was being given to some till the company realized the problem and took it away from the shelves in Jan 2025 @MoHFW_INDIA , Sir, please do issue an advisory. It's not enough that the fake version has been taken off the shelves. Those who have received the fake version need to be informed too, so that they take the vaccine again. The perpetrators of this crime need to be brought to justice!
47
710
31
1,673
98,018
In 2020, we did something very odd. Well, K (Kailash, CTO) did. He helped open-source Alar, a Kannada–English dictionary. It's a little absurd, considering we're a stockbroking company, but the project itself is one of monumental importance. The story of how Alar came to be is even more inspiring. It was essentially the life's work of one man: V. Krishna. Alar is the online version of what he had built over 40 years—researching, writing, and cataloguing more than 150,000 Kannada words and 240,000 English definitions, complete with all their attendant details. Just thinking about someone spending four decades relentlessly pursuing one single project is beyond inspiring. Oh, and he is still working on adding to the corpus. That @zerodha had even a small role to play in this is deeply gratifying. It has now been five years since Alar launched, and over two lakh people visit it every month. It also just received a major update, faster than ever. And K has been working on improving dictpress, the underlying open-source technology that powers the creation of online dictionaries includng Alar. So if, by any chance, you're obsessed with languages and dictionaries and have been wondering how to build one online, you should definitely check it out. (link in comments)
84
373
33
3,349
228,036
Digital Gold is unsafe and you may lose all your Saving parked into digital Gold🚨 🚨Not regulated by SEBI/RBI Digital gold sold by apps (PhonePe, Paytm, Groww, etc.) is not regulated, so there is no guarantee like mutual funds / banks. 🚨No legal claim if seller shuts down Your gold is stored by private vault partners. If the company goes bankrupt, recovery can be messy. 🚨Counterparty risk You rely on the seller + refiner + vault. If any one fails, you face risk. 🚨Holding limit only 5 years Most providers allow max 5 years storage. After that you must sell or convert to physical, which has extra charges. 🚨High buy–sell spread Digital gold prices have a 3–6% spread, making it costly compared to ETFs. 🚨All Charges (Short) 1. Making/Storage charges (indirect) included in buy price; usually 2–3%. 2. Buy–Sell Spread 3–6% difference between buying and selling price. 3. 3% GST on purchase (same as physical gold). 4. Delivery charges (if you convert to physical), Making charges + shipping charges apply. 5. Exit charges if vault storage expires. Some providers charge for storage beyond limit. Summary👇 Digital gold is risky mainly because it’s unregulated and depends entirely on private companies, plus it has hidden buy-sell spreads and storage-related costs.
145
589
47
2,451
389,498
Qwen doesn't get enough love and respect. They have been on 🔥
tonight we should have 6 releases
19
16
365
22,430
Tip: Never put large JSON data in the context of an AI conversation. This is the same query, but with the data formatted into a csv like format instead of json. Slashed costs by 20x. @mattpocockuk maybe you could do PSA about this since you have the reach.
73
71
13
1,544
254,375
I am happy I could contribute to this project, which led to an expected savings of over Rs. 200 crore for the market. The one thing we did beyond learning the existing systems and not being afraid to be wrong and correcting ourselves was to have 360 feedback (1/2)
One critical aspect of the brokerage business most people are unaware of is the sheer underlying complexity. At a high level, we integrate with multiple stock exchanges, depositories, RTAs, banks, and others. Each of these integrations (hundreds of APIs and file exchanges) has been non-uniform and non-standard, resulting in massive amounts of technical and operational complexity. These are completely invisible to a broker's clients. A great example of quiet, systemic backend work yielding significant impact is the multi-year exercise SEBI undertook to standardise data exchange across market entities—the Unified Distilled File Formats (UDiFF). The same data that was represented inconsistently across three exchanges, for instance, was unified to be consistent in a single format, across numerous data structures. As a result of this, we ended up removing 60%+ of the code that was required to maintain these separate data formats, significantly reducing technical and operational complexity and risk. Along with it, we saw huge efficiency gains. For instance, a nightly data import process went from taking 40 minutes to 30 seconds. Ultimately, this one SEBI project has cleared technical debt accrued over decades, significantly reducing technical and operational risk systemically in Indian capital markets. These boring, invisible, background updates are often far more meaningful, impactful, and directly beneficial to retail clients than most shiny frontend features. At @zerodha, we take technical and operational debt very seriously. Constantly reducing its buildup and giving these invisible improvements as much importance, if not more, as frontend features, is our engineering philosophy. Our priority is to make the underlying systems robust and reliable for the next decade, rather than focusing on what can be released in the next quarter.
1
15
from stakeholders at every step. And we asked a lot of questions! It's really exciting that after one year of its completion, it's started to show actual benefits. lnkd.in/dHPD6B57
7