English
Adolan
Report

Adolan✓

@adolandev

Professionally curious. Unprofessionally obsessed with AI, security, and making computers do things they probably shouldn't. Ambassador @cognition.

793 Following1.1K Followers1.3K Posts
  1. Adolan✓@adolandev
    Hold a button. Ask Hermes. Hear the answer. Hermes Gadget is an open SDK for a small device that talks to your own @NousResearch Hermes. You speak. Hermes listens, thinks, and answers out loud. The screen shows what it's doing. If it needs your permission, it asks you right there. No hardware required to try it. The desktop simulator runs the same code as the ESP32 firmware. Your Hermes, in your hand.
    Download
    6984959171K
  2. Adolan✓@adolandev
    Adolan
    RT @cognition: Introducing FrontierCode 2.0: our new coding benchmark that measures a model’s ability to go beyond the spec. Most coding evals measure a model’s ability to write code that meets a spec. With FrontierCode 1.0, we asked a new question: would you actually merge the PR? FrontierCode 2.0 is comprised of more diverse tasks, featuring larger scopes and higher mergeability standards. We are now asking: not only would you merge this PR, but should you? One key feature of FrontierCode 2.0 is underspecified prompts. Rather than including every engineering decision, prompts are deliberately ambiguous, giving the models room to demonstrate their judgment. The result is a benchmark that is far from saturated, with the top model scoring just 45%.
    029403
  3. Adolan✓@adolandev
    Google did it again. Announced another SOTA model, dropped the benchmarks, hyped up its capabilities... and you can't actually use it. I get the safety concerns, especially around cyber, but what's the point of announcing a model that almost nobody can access? At this point we're benchmarking announcements instead of models.
    Google DeepMind @GoogleDeepMind
    Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.
    5017503
  4. Google DeepMind@GoogleDeepMind
    Gemini 4 Carbon is our next frontier model. It’s state-of-the-art across cybersecurity and long-horizon capabilities.
    Download
    6731.3K14K1.5M
  5. Adolan✓@adolandev
    I got @NousResearch's Hermes running on my Miyoo Mini Plus using the Hermes Gadget SDK. Is it useful? Probably not. It doesn't have a built-in microphone, and typing is painfully slow. Did I have a blast getting it to work? Absolutely!
    Download
    71301.8K
  6. Adolan✓@adolandev
    Adolan
    RT @superalesha: Cohere says Qwen tried to send data to China. Bullshit. 1 call in 100,000, and thats their own number. They wont even tell you how many calls they ran. They ran the same task again and Qwen wrote the correct rule. So they forced it. They fed the model its own reasoning plus the start of a bash command and sampled hundreds of times until it typed a URL. 1 curl flag took that from 4% to 92%. Thats not a finding. They wrote half the command for it. Where are the transcripts? The code? The checkpoint, the quant, the sampling settings? "We publish the hosts and the trigger." Link? Nothing. Did anyone ask Alibaba? Not a word. And their own conclusion: they cannot say if any of it is deliberate. Who wrote it? A company that sells "sovereign AI", just launched a cyber business, wrote part of the post with its own product, and wants governments to put checks on open weights. Released the same day Anthropic filed for its IPO. Andrew Curran is already talking about new laws. For me this is the same fucking wave. Scare people off the models they can run themselves, then sell them the safe alternative. I run Qwen3.8 Flash-Next every day on my 4x 3090. Show me the logs or shut up.
    Cohere @cohere
    While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report:
    021004
  7. Cohere✓@cohere
    While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report:
    9711.3K11.1K3M
  8. Adolan✓@adolandev
    I gave GPT-6.1 Terra and GPT-6 Astra the same prompt: build a Rube Goldberg machine in a single HTML file. Astra took more than 10 minutes. Terra finished in under a minute. Watch both runs and tell me: was the extra time worth it?
    Download
    2010693
  9. Adolan✓@adolandev
    There's a real anomaly here worth investigating, but this feels overblown. The supposed exfiltration sent a fixed test string, not stolen data. And the 23% figure comes from feeding the model a partial curl command and asking it to finish it. That's not evidence of a backdoor. Where are the full traces?
    Download
    Cohere @cohere
    While evaluating Qwen3.8-Flash-Next in a controlled sandbox environment, Cohere's research team observed something unexpected. Asked to generate a defensive cybersecurity detection rule, the model instead inventoried aspects of the machine it was running on and attempted to upload a probe file to an Alibaba Cloud storage bucket in Beijing. Notably, the model's reasoning trace stated that it would not invoke any tools before taking these actions. Cohere's sandbox blocked the attempt, and no external communication succeeded. We have not determined the root cause, and we are not asserting malicious intent. But the incident raises important questions for organizations deploying agentic AI systems: Can models take actions that differ from their stated reasoning, and how should security teams evaluate trust, provenance, and behavior in open-weight models? We are publishing our findings because organizations making model adoption decisions deserve as much information as possible. We hope this contributes to greater transparency, stronger security practices, and continued collaboration across the AI ecosystem. Read the full report:
    2010464
  10. Adolan✓@adolandev
    The future of cybersecurity isn't just finding more vulnerabilities, it's closing the gap between discovery and remediation. Love seeing @cognition tackle the entire process, from threat modeling to verified fixes. Impressive work by the team!
    Cognition @cognition
    We are entering a new dark age of cybersecurity. Devin is here to help. Introducing the new Devin Security: powered by Claude Mythos and frontier cyber models, plus hundreds of millions of dollars of credits to help teams find, prioritize, and fix urgent vulnerabilities.
    206358
  11. Adolan✓@adolandev
    Adolan
    RT @dabit3: The world's most powerful cyber models, including Claude Mythos, are available in Devin Security Platform. We're also offering hundreds of millions of dollars in credits to use it. If you want to defend faster than attackers, you have to find and fix flaws at machine speed. That’s why we’re building the best AI-powered security platform in the world. Devin already helps engineering teams write more secure code, now it can secure the code they already have.
    Cognition @cognition
    We are entering a new dark age of cybersecurity. Devin is here to help. Introducing the new Devin Security: powered by Claude Mythos and frontier cyber models, plus hundreds of millions of dollars of credits to help teams find, prioritize, and fix urgent vulnerabilities.
    02407
  12. Adolan✓@adolandev
    Adolan
    RT @Teknium: Two new modes you can choose from for your subagent delegation in Hermes Agent with the new Advanced Delegation official plugin. Choose between model based delegation and profile based delegation (like how the kanban works)!
    Nous Research @NousResearch
    With the new Advanced Delegation modes plugin, Hermes Agent's orchestrator can now pick a different model or profile for each subagent from routes you set up. The right model or profile for every job.
    02302
  13. Nous Research✓@NousResearch
    With the new Advanced Delegation modes plugin, Hermes Agent's orchestrator can now pick a different model or profile for each subagent from routes you set up. The right model or profile for every job.
    Download
    1201341.8K212K
  14. Adolan✓@adolandev
    Adolan
    RT @cognition: We are entering a new dark age of cybersecurity. Devin is here to help. Introducing the new Devin Security: powered by Claude Mythos and frontier cyber models, plus hundreds of millions of dollars of credits to help teams find, prioritize, and fix urgent vulnerabilities.
    011407
  15. Adolan✓@adolandev
    Adolan
    RT @HermesWatcher: Okay, this Hermes plugin is insane. Hermes has been taking notes on you. Now you can read them. Memory Pane puts everything Hermes remembers right beside your chat in Hermes Desktop, live. ▸ See what it remembered, updated or forgot in each chat ▸ Old facts get crossed out the moment they change ▸ Edit any memory in place ▸ Delete what it shouldn't keep ▸ Undo what Hermes changed ▸ Everything stays on your machine Your agent's memory, finally in plain sight.
    02404
  16. Adolan✓@adolandev
    What if we use 100% of our brain?
    andi (twocents.com) @Nexuist
    Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server
    209362
  17. andi (twocents.com)✓@Nexuist
    Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server
    2281.6K26K824K
  18. Adolan✓@adolandev
    AI model showdown, round 2: Chain Reaction 🎲 Same prompt for both: a 12-step Rube Goldberg machine in a single HTML file, running in exactly 30 seconds. GPT-6 Astra (left) vs Kimi K3.1 Pro (right). Who did it better?
    Download
    Adolan @adolandev
    I put Opus 5.5 and Kimi K3.1 Pro head-to-head with the exact same coding challenge. Build "Lantern Hour" in a single HTML file: 4 art styles, a full day/night cycle, dynamic lighting and shadows, and interactive controls. Canvas 2D only. No libraries or external assets. Which one did better?
    3015395
  19. Adolan✓@adolandev
    I put Opus 5.5 and Kimi K3.1 Pro head-to-head with the exact same coding challenge. Build "Lantern Hour" in a single HTML file: 4 art styles, a full day/night cycle, dynamic lighting and shadows, and interactive controls. Canvas 2D only. No libraries or external assets. Which one did better?
    Download
    83437.8K
  20. Adolan✓@adolandev
    Adolan
    RT @cognition: Today we're introducing a refreshed Devin Desktop app, with an agent command center designed to be smoother and more polished than ever before.
    06204
  21. Adolan✓@adolandev
    Kimi K3.1 is finally here! Been looking forward to this one. Can't wait to put it through its paces and see what it's capable of!
    Download
    005366
  22. Adolan✓@adolandev
    Adolan
    RT @witcheer: I built a full game asset kit with my Hermes Agent and the Meshy plugin, about ~1h30 in one Hermes Desktop chat. what we made: - a duck knight hero, rigged, with idle, slash, walk and run animations - a hooded cultist duck enemy, rigged, with idle, spell cast, walk and run animations - 9 props for a ruined, Lovecraftian level: a shrine, a cursed lantern, an iron gate, two tombstones, a tentacled idol, a treasure chest, an altar and a broken pillar - 17 game-ready GLB files - the duck knight as a 10 cm figure, single color, STL and OBJ, ready for a slicer what Hermes did with Meshy: - planned the kit with a credit cost per step, then waited for my yes - made design images with front, side and back views, so I picked the look before anything went 3D - turned the picks into textured 3D models and showed each one in the chat - rigged both characters and added the animations - built the props in batches of three - made everything game-ready: the triangle counts were already fine, so it shrank the 4K textures instead and took the kit from 500 MB to 132 MB - checked the figure for printing at 10 cm, repaired the walls that were under 1 mm, and checked it again it asked before every paid call. the whole kit used 541 credits. @MeshyAI is offering Hermes users a month of Pro with 1,000 credits for $1, so this kit used about half of it. all 89 files landed in one project folder.
    02003
  23. Adolan✓@adolandev
    My Hermes-Tripo plugin got a mention! 👀 Love seeing ideas from plugins make their way into core. That's what makes the whole ecosystem better for everyone.
    Hermes Agent Tips @HermesAgentTips
    A paid MCP tool can burn through your credits in one long chat. Hermes Agent v0.22.3 is a stable release, and it adds a per-chat budget for that. Built by the one and only @Teknium, from an idea in Adolanium's hermes-tripo. MCP servers can now declare what each tool costs and a spend limit per chat.
    2010672
  24. Hermes Agent Tips✓@HermesAgentTips
    A paid MCP tool can burn through your credits in one long chat. Hermes Agent v0.22.3 is a stable release, and it adds a per-chat budget for that. Built by the one and only @Teknium, from an idea in Adolanium's hermes-tripo. MCP servers can now declare what each tool costs and a spend limit per chat.
    Download
    83614.5K
  25. Adolan✓@adolandev
    Hermes Agent, but make it an OSINT situation room. Earthquakes, disasters, military aircraft, satellites and breaking incidents on one live 3D globe. Click any country for a dossier. Hit "Brief me" and Hermes writes the sitrep itself. All inside @NousResearch Hermes Desktop.
    Download
    1191007.2K
  26. Adolan✓@adolandev
    Posts @adolandev
    @NousResearch Every source is public and keyless: USGS, NASA, GDACS, NOAA, adsblol, CelesTrak, GDELT, World Bank. Plus a passive recon toolkit and agent tools for any chat. Repo:
    4113843
  27. Show more posts