AICharsiya!
5.9K posts
AICharsiya!
@AICharsiya
Infosec Professional | Geek | Digital Art Creator | Photographer | AI Enthusiast
India Katılım Mayıs 2009
1.2K Takip Edilen491 Takipçiler
Assam and Bengal are clear cases of the election being stolen by the BJP with the support of the EC.
We agree with Mamata ji. More than 100 seats were stolen in Bengal.
We have seen this playbook before:
Madhya Pradesh.
Haryana.
Maharashtra.
Lok Sabha 2024 etc
चुनाव चोरी, संस्था चोरी - अब और चारा ही क्या है!
English
I did it
I generated 100M tokens on @OpenRouter for free
All thanks to Owl-Alpha
For reference Opus 4.7 costs 5$/M input tokens and 25$/M output tokens
Let’s assume 100% of tokens generated were output tokens
Total cost for 100M tokens would’ve been 2500$
In 24hrs
English
@ivanfioravanti @pupposandro @davideciffa My dyslexia ass read it as lubebox team! BRB washing me eyes
GIF
English
AICharsiya! retweetledi
Replit Agent will be FREE for 24 hours on Saturday. 🤯🤯🤯
Replit ⠕@Replit
Free Agent usage starts at 5:00am PST on May 2 for 24 hours
English
Equal credit to NVIDIA, Amy Steier, Andre Manoel, Alexa Haushalter, and Maarten Van Segbroeck, for shipping the Nemotron-PII dataset via NeMo Data Designer + Nemotron-Personas.
Persona grounding is what made it realistic enough to fine-tune on.
English
Same text. Two privacy filters.
OpenAI's model catches 8 categories. OpenMed catches 55+: medical record numbers, blood type, API keys, financial codes, demographics.
Trained on Nemotron data by Nvidia. All on-device. All open-source.
Coming soon! What's missing?
English
The conclusions here feel wrong to me. The two lessons I see are:
1. Don't run agents anywhere they might be able to access production environment credentials - it's on you to know which credentials those are
2. Keep tested backups that are independent from your production host
JER@lifeof_jer
English
English
The DeepSeek V4 garbled output bug in open source inference engine is fixed in SGLang.
To everyone affected over the weekend, sorry for the trouble.
Huge thanks to @Ant_Group for landing the fix PR. It was a cross-company, cross-timezone, sub-48-hour marathon. @ollama and @humansand surfaced it first; @nvidia, @AIatMeta, and @FireworksAI_HQ raised the same signal soon after. @deepseek_ai replied in seconds at every hour. @FireworksAI_HQ stayed up late with us until it shipped. @SemiAnalysis_ and @ollama provided the machines that made the debugging possible. The SGLang team dug in through the weekend.
The real OSS is the friends we made along the way.🫶
English
English
@Prince_Canuma @pcuenq Nice! Unfortunately I can’t run any of them on a 96G vram laptop! :(
English
DeepSeek-v4 now runs at ~23-26 tok/s on MLX!
I made some custom kernels for the sinkhorn and it took gen speeds for 17 -> 26 tok/s.
The weights are also significantly smaller thanks to @pcuenq tip about keeping the experts in MXFP4!
Now you can use it to power your local coding agents (PI, Open code, Hermes agent or even CC)
PR:
github.com/ml-explore/mlx…
Prince Canuma@Prince_Canuma
DeepSeek-V4 running on M3 Ultra 🚀 Don't mind the speed, that's gonna improve soon.
English
GPT 5.5 is definitely a step up in the character game.
English
AICharsiya! retweetledi
Competition between Chinese labs is intensifying. Top 3 Open Models in Text Arena are all now sitting just below the top proprietary tier, each leading different real-world categories.
All category rankings are based on overall, which includes proprietary models.
#1 open GLM-5.1 by @Zai_org (#15 overall), the leading open-source model
- #10 Life, Physical & Social Science
- #10 Longer Query
- #12 Math.
#2 open DeepSeek-v4-Pro by @DeepSeek_AI debuts (#20 overall)
- #1 Medical & Healthcare
- #15 Creative Writing
- #18 Multi-Turn
DeepSeek’s V4 Pro (Thinking) variant also excels in:
- #8 Math
- #9 Life, Physical & Social Science
#3 open Kimi-K2.6 by @Kimi_Moonshot (#26 Overall)
- #6 Chinese
- #10 Instruction Following
- #17 Expert
Arena.ai@arena
Exciting news - DeepSeek V4 Pro is in the Arena with 1.6T parameters (49B activated) alongside V4 Flash at 284B parameters (13B activated). Both support 1M token context. It’s a major leap over DeepSeek V3.2! Code Arena: - DeepSeek V4 Pro (thinking): #3 open model (#14 overall), on par with GPT-5.4-high and Gemini-3.1-Pro in agentic webdev tasks Text Arena: - DeepSeek V4 Pro (thinking): #2 open model (#14 overall), matching Kimi-2.6 - DeepSeek V4 Flash (thinking): #10 open model (#47 overall) Competition at the top of the open model leaderboards keeps heating up. Huge congrats to @DeepSeek_AI on the strong comeback!
English
AICharsiya! retweetledi
💥 I Ran DeepSeek V4 Flash Locally and Got Blown Away
♠ and this is the closest thing to a datacenter in your terminal
🔹 284B parameter model running on 2x H100s — fully local, zero API calls
🔹 FP4 + FP8 mixed precision — fits in 192GB VRAM with room to breathe
🔹 1M token context window — your entire codebase, novel, or knowledge graph in one shot
🔹 Matches GPT-5.4 on Codeforces, beats it on reasoning — open source MIT license
🔹 Full setup walkthrough — NVIDIA Brev, vLLM, DeepSeek native inference, the works
🔥 Full video below 👇
English
@ylecun @RpsAgainstTrump But if he wins then isn't his assumption correct?
English
@RpsAgainstTrump Dumbest president ever who also assumes people are even dumber than him.
English
Trump: “I took a lot of heat for saying drugs were going down 500%, 600%, 700%. But we also say sometimes 50%, 60%, it’s a different kind of calculation, and people understand it better.”
The dumbest president ever.
English
AICharsiya! retweetledi
Let's go DeepSeek v4!!! 🚗🚀
LMSYS Org@lmsysorg
DeepSeek V4 by @deepseek_ai just dropped! SGLang is ready on Day 0 with a full stack of optimizations from architectures to low-level kernels. We also deliver a verified RL training pipeline in Miles (by @radixark) for V4 at launch: 1️⃣ Native "ShadowRadix" Design: DeepSeek V4's hybrid attention is complex. Our new ShadowRadix engine is the first to provide native prefix caching for SWA and compressed KV pools, making 1M+ context retrieval seamless and memory-efficient. 2️⃣ High-Performance Kernels: - Flash Compressor: IO-aware fused kernels, 10x faster than naive implementations. - Lightning TopK: High-speed indexing for 1M context in just 15µs. - Integrate FlashInfer trtllm-gen MoE, FlashMLA, and MegaMoE kernels 3️⃣ Rich Features: Speculative decoding, HiSparse, Attention DP/TP/CP and MoE TP/EP, and multi-platform support 4️⃣ Verified RL: The open-source RL pipeline: full parallelism (DP/TP/EP/PP/CP), tilelang kernels, tensor-level checked precision, verified with growing reward. Get started immediately with our out-of-the-box Cookbook 👇 Enjoy! #DeepSeekV4 #SGLang #LLM
English
AICharsiya! retweetledi
Gemini Embedding 2 is now generally available in the Gemini API and Vertex AI!
Start building with our first natively multimodal embedding model, now equipped with the stability and optimizations required for production apps.
English
AICharsiya! retweetledi
#Gurgaon #DLF1
illegal ramps and encroachments cleared to restore public spaces. Action on ground, zero tolerance.
#DTPE #AmitMadholia #DLF1 #DemolitionDrive #EncroachmentFree #LawForAll #UrbanPlanning #Gurgaon
English
AICharsiya! retweetledi
Why is no one talking about this?
@nvidia is offering around 80 AI models via hosted APIs absolutely for free.
You get access to MiniMax M2.7, GLM 5.1, Kimi 2.5, DeepSeek 3.2, GPT-OSS-120B, Sarvam-M etc.
This plugs straight into OpenClaude, OpenCode, Zed IDE, Hermes agent and even with Cursor IDE.
Setup:
– Grab API key: build.nvidia.com/models
– base_url = "integrate.api.nvidia.com/v1"
– api_key = "$NVIDIA_API_KEY"
– select model (e.g. minimaxai/minimax-m2.7)
If you’re building or experimenting, this is basically free inference.
Lock in and start building today anon.
Thank me later.
English
@CrimeBranchDP 20k is the problem. No one would snitch on that low amount. Offer more cash and may be 14 Years will be reduced to 14 months or even lower. Congratulations nonetheless! 🎉👏
English
🚨⚡ 14 YEARS ON THE RUN… ₹ 20,000/- REWARDEE FINALLY CAUGHT! DELHI POLICE STRIKES BIG 🚔🔥
👮♂️In a major breakthrough, CR, Crime Branch arrested a dreaded ₹20,000 rewardee Proclaimed Offender who had been absconding for 14 years in a robbery-cum-abduction case of Dwarka South.
🔴 Evaded arrest by changing cities, mobiles & identity
🔍The accused is linked to 13 criminal cases including robbery, burglary, theft, Arms Act & more.
💪 Successful operation led by Inspr Mahipal Singh under ACP Satendra Mohan and overall supervision of DCP Sh. Sanjeev Kumar Yadav
🚨⚡ 14 साल से फरार… ₹20,000 का इनामी बदमाश आखिरकार गिरफ्तार! दिल्ली पुलिस का बड़ा वार 🚔🔥
👮♂️ एक बड़ी सफलता में CR, क्राइम ब्रांच ने ₹20,000 के इनामी घोषित अपराधी (Proclaimed Offender) को गिरफ्तार किया, जो द्वारका साउथ के robbery-cum-abduction केस में पिछले 14 वर्षों से फरार चल रहा था।
🔴 गिरफ्तारी से बचने के लिए लगातार शहर बदलता रहा, मोबाइल नंबर बदलता रहा और पहचान छुपाता रहा।
🔍 आरोपी के खिलाफ डकैती, चोरी, सेंधमारी, आर्म्स एक्ट समेत 13 आपराधिक मामले दर्ज हैं।
💪 यह सफल कार्रवाई इंस्पेक्टर महिपाल सिंह के नेतृत्व में, एसीपी सतेंद्र मोहन के मार्गदर्शन तथा डीसीपी श्री संजीव कुमार यादव की समग्र निगरानी में की गई।
shorturl.at/VvXiZ
#DelhiPolice
#CrimeBranch 🚔
@delhipolice
English
@MiniMax_AI I use Mindcraft earlier but the AI player was cuckoo 🤪. How does visual feedback work here i.e. how does the model see what's on screen?
English
8 agents. 1 city. this is what M2.7's long-horizon tool use was built for 🔥
Alexander Knigge@AlexanderKnigge
Fire market stall, bro. @Minimax @MiniMaxAgent M2.7 playing Minecraft I hooked up 8 agents to a custom Mineflayer agent harness running M2.7 and told them to build a city. Ember here is building a market stall #minimax
English
AICharsiya! retweetledi
I’m hearing there’s renewed lobbying in DC and in state legislatures to ban or severely restrict open-source.
Like a few years ago, we’ll need everyone to help show policymakers why open-source matters: for startups, for competition, for economic growth, and for jobs.
If you build with open-source, now is the time to speak up!
English
AICharsiya! retweetledi
Keep improving 🚀🚀 @arena
Arena.ai@arena
Qwen3.6 Plus lands at #7 in Code Arena with a score of 1476 - up +16 points since the Preview. The new score also moves @AlibabaGroup to #3 lab in Code Arena. In the Text Arena, Qwen3.6 Plus lands at #36, a +13 point improvement since Preview. Congrats to the Qwen team on the continued progress!
English