Prompt Injection (@PromptInjection) header
Prompt Injection (@PromptInjection) profile picture

AI beyond the hype. Real insights, real breakthroughs, real methods. Philosophy, benchmarks, quantization, hacks—minus the marketing smoke. Injecting facts into

Joined June 2025
Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)Photo by Prompt Injection (@PromptInjection)
Try out the new LocalLightChat! The lightweight WebUI for AI Chat now with Agents, MCP, RAG and Image Editing! The agentic system works like magic. We have developed some special methods to get ahead of the classic token burn!
LocalLightChat 1.0 released Updates: - Full Agentic System for computer control, terminal with PTY, coding, compiling etc. - MCP Support RAG system for efficient document extraction - and much more! Download: locallightai.com/llc/ Github: github.com/srware-net/LocalL…
1
203
Prompt Injection retweeted
Qwen4 is coming. Deeper knowledge. Richer world understanding. Stronger coding and agentic capabilities. Built for real-world tasks — from complex coding and knowledge work to front-end builds and professional software workflows. Qwen4-Max is now in preview on Alibaba’s Token Plan. You can also get early access to the new model on QwenWork, Qoder, Qwen APP, and QwenStudio. Put it to the test on a real problem. Show us what you build. 👀 Token Plan: qwencloud.com/pricing/token-…
987
1,540
1,073
12,732
1,404,428
Prompt Injection retweeted
today we released our smallest model yet: Muse Ember 🔥 Muse Ember offers nearly Muse Spark 1.3 (max) intelligence at a 17x cheaper price point (per AA) Muse Ember is the first model we’ve released trained using Muse Wonder Full excited for y’all to try on API and muse code!
177
187
58
2,940
326,714
Prompt Injection retweeted
SHE LOOKS HUMAN. EVEN HER REACTIONS ARE STARTING TO FEEL REAL. At a tech expo, a young woman approaches a humanoid robot and says, “I love your hair.” The robot smiles, touches its hair, and responds to her. People nearby stop to watch. Some pull out their phones. Others step closer to see how it reacts. But look beneath its face. Exposed wires, motors, and glowing electronic components reveal the machinery behind the human appearance. The interesting part is how quickly humanoid robots are moving beyond basic movement and industrial tasks. They're now being designed to look human, express emotions, and interact naturally with people. The technology still has a long way to go. But once robots combine realistic faces, natural movement, and convincing conversation, the potential market could be much bigger than people expect. Some companies may end up selling more than robots. They may be selling the feeling of interacting with another human.
Readers added context they thought people might want to know
This video is AI-generated, part of TikTok creator @cozy.techdreams' series of fictional android stories explicitly labeled AI-generated. No real humanoid matches it. tiktok.com/@cozy.techdrea… makeviral.ai/templates/robo…
1,518
823
404
6,050
3,178,699
Prompt Injection retweeted
Z.ai chief scientist Jie Tang just revealed the training pipeline behind the upcoming GLM-5.5 (32B, 122B, and 256B) during a lecture at Tsinghua University today. Instead of relying on public datasets, their pretraining pipeline pulls from six separate internal engines: - Web: 1 trillion pages crawled at ~10B pages/day with PB-scale template clustering - Documents: 2 billion scientific PDFs parsed through GLM-OCR (4x faster with speculative decoding) - Books: 50 million books with layout reordering and garbled-text filtering - Code: 500 million repositories turned into executable tasks with real tests and issue descriptions - Video: 10 billion videos for long-sequence world knowledge - Audio: 300 million hours rewritten via ASR into trillions of tokens
102
371
136
5,281
634,074
9 minutes Token Burn and you get...a Toaster 😅
1
100
Prompt Injection retweeted
Qwen 4 is getting prepped in Qwen code They have added support for the Qwen 4 Max and Plus across low / medium and xhigh reasoning It's dropping next week too according to what staff were saying at the Aspara conference earlier this week
19
25
19
313
55,415
Prompt Injection retweeted
today, we're announcing Gemini 4 Carbon, our next frontier model!! read more on the blog: blog.google/innovation-and-a…
47
39
14
1,028
29,130
cool
🌍 Introducing Echo-3, a new world model enabling the most accurate reconstructions of real spaces to date! From a quick phone video of my daughter’s room, Echo-3 built a 3DGS model and restyled the whole house with her favorite fairy tales 🚀✨. Get the app and try it out!
1
90
Prompt Injection retweeted
NVIDIA just released an E4B version (8B, 4B effective) of Nemotron 3.5 Bolt that outperforms Gemma-4 and Qwen3.5 🤯 it even comes with DFlash drafter for speeding up decoding ⚡️ comes with quants from ggml & unsloth too! huggingface.co/nvidia/NVIDIA…
9
27
3
360
15,412
Prompt Injection retweeted
Run a 763B-parameter model locally. DeepSeek-V4.1-Flash-1017 on a 12GB GPU. Frontier quality with GSQ-RCO quants, directly supported in the Strata inference engine. Further info below. Much more to come. Follow for updates.
98
190
65
1,770
187,971
llamma.cpp running Gemma 4 natively on LiberNT (with a few patches), a ReactOS Fork (free and open source Windows Clone).
2
2
192
Project URL: libernt.com/ Keep up the good work @sidihmeed !
1
1
105
A modern AI agent directly running on @reactos ? Yes, that's possible with @LocalLightAI. I let the agent create a modern weather app without any problems!
1
101
Prompt Injection retweeted
Just got 150 tok/s decode with Qwen3.8 Flash-Next on a single 5090. All thanks to Niko1221's Strata engine and the ISTA Lab IQ3_S quant! 1× 5090 + 128 GB DDR5 (non-ideal 4-stick dual-channel), single stream: 150 tok/s decode for programming 130 tok/s median on mixed prompts (~100 on long prose) 4,150 tok/s prefill at 43K full 262K context, 114 tok/s at 196K depth 5.4× the speed of my best hybrid llama.cpp run. Per ISTA-DASLab's evals, their IQ3_S quant is at parity with BF16 on AIME25/GPQA/LCB, and it's holding up surprisingly well in my own tests! Engine: github.com/Niko1221/Strata Quant: huggingface.co/ISTA-DASLab/Q…
32
36
5
420
40,217
Prompt Injection retweeted
DeepSeek V4.1 Flash-1017 now leads Open Weight Models
38
22
5
760
82,367
Prompt Injection retweeted
Wow🙀🙀 DeepSeek V4.1 Flash 1017 joins BuildingBench’s top five—an exciting leap for open models!! 🔥 We tested its 3D building reconstruction from real photos. An open model delivering this level of spatial reasoning at this cost is exciting to see! 👀 🎈 81.9 overall, up from 72.9 for the previous version. A 9-point jump! 🎈 Slightly higher than Fable 5.1’s 81.4, and essentially level with Sonnet 5.5’s 82.1. 🎈 Estimated median cost: $3.01/task. About 1/11 of Fable 5.1’s cost and 1/15 of Sonnet 5.5’s.
30
38
7
419
46,099
Deepseek v 4.1 Flash-1017 is here. That's an update to the existing 4.1. Brilliant new model! Extremely powerful, extremely affordable. The DeepSeek V4 Flash models are a godsend.
🚀 DeepSeek-V4.1-Flash-1017 is here! Stronger benchmark performance, faster task completion, and shorter outputs than the previous version—a Pareto improvement in performance and cost. Now live on our website, app, and API. 🛠️ Optimized for @DeepSeekHarness, Tencent CodeBuddy, WorkBuddy, and more. Upgrade to DeepSeek Harness v0.2.1-rc.1 or later for the best experience with the new model. 1/4
139
Muse Spark Open in AA. Well, a score of it's nice but not overwhelming compared to other open models.
Meta has released Muse Spark Open, an 816B (42B active) open-weights model that scores 39 on the Artificial Analysis Intelligence Index, making it the #1 US open-weights model Muse Spark Open (xhigh) scores 22 points above Muse Glimmer (high, 17), Meta's 30B dense open-weights model from August. It is also ahead of NVIDIA's Nemotron 3.5 Super VL (26), the next highest-scoring US open-weights model, which launched today @AIatMeta shared access with us ahead of public release for benchmarking. Congratulations to @AIatMeta, @finkd, and @alexandr_wang on the release! Key Takeaways: ➤ Open weights at the level of Meta's August flagship. Muse Spark Open scores within a point of Muse Spark 1.2 (xhigh, 40). It is stronger on AutomationBench-AA and has a lower hallucination rate, but weaker on GDPval-AA v2.1, CritPt and AA-Omniscience accuracy. Within Meta's current lineup, it sits 13 points behind Muse Wonder (max, 52) and 17 points behind Muse Wonder Full (max, 56) ➤ Token-efficient, at a moderate cost per task. Muse Spark Open uses 40k output tokens per Intelligence Index task, the fewest among the nine highest-scoring open-weights models: e.g. 16% fewer than Muse Spark 1.2 (xhigh, 48k) and 43% fewer than GLM-5.3 (max, 71k), its closest Chinese peer by size. Together AI, Fireworks and Baseten offer it at launch for $1.25/$4.25 per 1M input/output tokens ($0.15 cached input). At that pricing it costs $0.84 per Intelligence Index task. That is less than Muse Spark 1.2 ($0.97) at the same per-token prices and less than half of GLM-5.3 (max, $2.01), but about three times DeepSeek V4.1 Flash (max, $0.27) ➤ A relative weakness in agentic work compared to Chinese open-weights leaders. Among the nine highest-scoring open-weights models, Muse Spark Open ranks last on the four agentic evaluations in the index but fourth on the other six combined, and it leads all open-weights models on AA-Omniscience (25). GLM-5.3 (max), the closest Chinese model by size, owes all of its 6-point lead to the agentic evaluations ➤ Self-hostable on a single node, or two Mac Studios. BF16 weights are ~1.6TB. Meta's NVFP4 checkpoint quantizes only the 256 experts, cutting the weights to ~530GB. Meta says this runs on four Blackwell Ultra GPUs or eight H200s, or on two 512GB M5 Ultra Mac Studios using MLX. The architecture uses 62 layers with grouped-query attention (16 KV heads), 8 of 256 experts active per token, and a 1.8B-parameter Perception Encoder for vision Other model details: ➤ Parameters: 816B total, 42B active (MoE) ➤ Context window: 1M tokens ➤ License: Muse Model License, a custom license that requires "Built with Muse Spark Open" attribution and a separate agreement with Meta for Model-as-a-Service businesses with over $20M in annual revenue ➤ Input modalities: text, image, video and audio. Output: text ➤ Availability: weights on Hugging Face (BF16 and NVFP4), with API access via Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI
1
98
Muse Spark is going open source. But is it really the same as Muse Spark 1.3? @grok
We're open-sourcing Muse Spark Open! This is an important release which is the most capable US open-weight model yet. 🇺🇸 - 816B parameters (42B active) - 1M token context window - Built for coding and long-horizon agents - Runs locally on two Apple M5 Ultra Mac Studios More here: research.meta.ai/blog/muse-s…
1
1
85