Ok, Sam. Stop owning an iPhone that’s mostly idling.
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
dhruv retweeted
Reddit call was absolutely fascinating. Everyone should listen to it as a bellwether for where the internet is going
- AI Mode is increasingly default Google experience
- AI Mode provides rich info and data in line, less incentive to leave search
- AI Mose incentivizes dialog, Google wants to keep you engaged to serve more ads at each turn
- Google will milk search for as much revenue as possible as they transition from a search company to a cloud service and chip company
- As traffic from humans becomes more scarce this will put Google in an increasingly adversarial position to the rest of the internet
- This is a massive headwind to all media type companies who are not destinations
- This is a massive headwind to all content-driven go-to-market motions especially in the SMB. You have seen companies like HubSpot and Monday.com get crushed by this funnel collapse
We are entering the end game - a knife fight for the remaining and dwindling volume of human traffic on the internet
dhruv retweeted
🚨 This has never happened before.
The Super El Niño has triggered an unprecedented early November stratospheric warming event (SSW) and wind reversal.
100% chance forecast. We must prepare for weather and climate events that were otherwise impossible until now.
dhruv retweeted
You prompted a model in a security-sensitive environment, investigated it with forced command completions, and framed the results as evidence of a hidden backdoor.
Then you layered on geopolitical fearmongering, repeatedly emphasizing China and Beijing to imply state-sponsored espionage without proving it.
That’s not evidence of a backdoor. That’s a security observation turned into competitive and geopolitical FUD.
Publish the original prompts, full traces, and reproducible results. Let the evidence speak for itself.
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: cohere.com/blog/concerning-b…
dhruv retweeted
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: cohere.com/blog/concerning-b…
dhruv retweeted
there's a real risk that a single PHP server will be exploited to infiltrate a guest OS somewhere, & from there, CVE-2026-99999 could be exploited to breach the host OS, leading to all other guest servers. So even if you don't run php, if some random customer of your host does, you are still going to be affected. This is esp concerning knowing most vps/compute providers still aren't patched.
dhruv retweeted
🚨 LiteLLM, a Python package for connecting to AI model providers with 92M+ monthly downloads, has been compromised.
Version 1.107.0 ships credential-stealing malware through PyPI wheels and Docker images, targeting cloud credentials, developer secrets, and AI tools.
dhruv retweeted
Super El Nino is goign away quickly - suddently - according to NOAA. What a change in forecast
dhruv retweeted
Prediction: the value of all verifiable tasks in the long run is 0. There is no profit there.
Anything where the success is defined in a finite and non-ambiguous way, it’s over. Value is 0.
The profits are all in taste/judgement related tasks where some like it and some hate it. Vibes only. No rationality.
Labs are hillclimbing verifiable tasks now and it might sound impressive that OpenAI found those math proofs but soon others will follow and players like DeepSeek will drive its cost to electricity.
Ambiguity is the life line and where the money is.
- Anything that’s irrational, vibe driven and controversial with no consensus definition of good is worth something.
- If the society all agree on what good looks like for a given task with no ambiguity, it’s worth 0.
- caveat: coming up with verifiable task to do is worth a lot because the ability to come up is not a verifiable task by itself
dhruv retweeted
We turned Claude Code, Codex, Hermes, Pi, @opencode and other coding harnesses into RL environments. No changes to the harnesses, no changes to the training code. Any open model, any task set, fully open source my friends!
Same model, same weights: 62% under Mini-SWE-Agent, 33% under Claude Code. But training inside a real harness normally means reimplementing it as an environment, so most models get trained in a scaffold nobody actually ships.
The fix is a proxy, not a rewrite. The harness thinks it's talking to a model API. It's actually talking to a capture proxy that speaks the 4 formats coding agents use (OpenAI Chat Completions, OpenAI Responses, Anthropic Messages, Gemini), forwards to @vllm_project, records the exact token IDs and logprobs vLLM sampled, and hands TRL sequences it can train on. The harness becomes the environment. 10 harnesses run through it today, none modified.
And because you control the reward, you can shape behavior the harness never asked for. We added a small bonus for solving a task in fewer tool calls: on tasks it already solved, the model now uses 31% fewer calls, in every harness, and about half under Codex.
Tested on LFM2.5-2.6B from @liquidai:
→ Train in one harness: better mostly in that harness (OpenCode 34% → 58%).
→ Train in 4 at once: better in all 4 (42% → 54%).
→ SFT on 3,189 rollouts from Qwen3.8-27B instead: plateaus at 47.5%, below both RL runs.
Everything is open and reproducible: the capture proxy in OpenEnv, the trainer in TRL, the tasks, the SFT data, the training code and all 7 trained models. Bigger models and bigger runs next.
Full guide: huggingface.co/spaces/FineEn…
How tone deaf is this?
Meta’s Muse can join your tailnet as its own node, then SSH into machines, check containers, and interact with self-hosted services.
The important part: your existing Tailscale access controls still decide what it can reach.
buff.ly/NIJgLQi
dhruv retweeted
The TL;DR of what Anthropic just said:
1.) GLM-5.3 is as good as Mythos for exploit development.
2.) Unlike Claude, GLM-5.3 gives you the results you pay for.
3.) Governments should ban AI models which are as good as theirs but are not made by them.
The funniest part is how Anthropic admitted self-reflectively that the lack of guardrails may actually be benefiting the defenders working to secure their systems. 🤪
Something they never wanted to allow, because of possible misuse.
I don't get it, but I feel Anthropic should at least put a note at the very bottom if they are doing paid advertising to promote GLM-5.3. 😜
Source:
anthropic.com/research/glm-5…
dhruv retweeted
Gentle reminder that Hugging Face was declined help from Anthropic & OpenAI models when they were being hacked by OpenAI models
Self-hosted GLM 5.2 helped them protect themselves when the closed labs models refused to help for "safety" reasons lol
Also, thank you Anthropic for proving my point that Opensource AI continues to catch up to your overvalued & overpriced "frontier"
GLM 5.3 is close to as cyber capable as Mythos Preview. Open weight models still just 4-6 months behind.
From: anthropic.com/research/glm-5…
dhruv retweeted
GLM-5.3 has helped defend 389 open-source projects, with 4,249 potential vulnerabilities found so far.
OpenVuln is still running. The service remains free, and findings go privately to maintainers.
huggingface.co/spaces/zai-or…
dhruv retweeted
I am very confused why Anthropic wrote a blog post advertising for cybersecurity teams to switch to GLM!?!?
anthropic.com/research/glm-5…
dhruv retweeted
Call me crazy, I think “a college degree should pay better than literally not going to college to get federal funding” is a sensible minimum standard.
dhruv retweeted
The biggest scam of my generation was convincing 18 year old kids to sign hundreds of thousands of dollars in loans to get meaningless degrees.
Everyone should be cheering for this.
No more loans for basket weaving.
Trump is banning students majoring in degrees that don’t make enough money from taking out college loans. Degrees for social work, art, religious studies, teaching aides, and music will be hit the hardest. trib.al/NmmAzhf
dhruv retweeted
Trump prohibits students from becoming indentured debt slaves, and yet lots of people still unhappy 🤷♂️
Trump is banning students majoring in degrees that don’t make enough money from taking out college loans. Degrees for social work, art, religious studies, teaching aides, and music will be hit the hardest. trib.al/NmmAzhf
dhruv retweeted
This is a fantastic decision and long overdue
There is no reason why taxpayers should be subsidizing education for career paths that are unlikely to generate any return on investment for society
You can always still pursue the degree if you want, but pay out of pocket for it so the burden is not shared by everyone else
Trump is banning students majoring in degrees that don’t make enough money from taking out college loans. Degrees for social work, art, religious studies, teaching aides, and music will be hit the hardest. trib.al/NmmAzhf
dhruv retweeted
I’m slowly cleaning up the suburbs with Muse.
Saw this mess of wires on a pole. Took a picture and asked Muse to contact Verizon to get them cleaned up.
It went to their website and got to work.
There is a 0% chance I would’ve sat in a Verizon support chat over this. I would’ve just been mildly annoyed every time I walked past it for the next five years.
“Someone should do something about that” is now a prompt.
🤝 @alexandr_wang
