sathish316

6 ms·

31.
by sathish316 7mo ago
Cowork Chat. Anthropic can do this. What is wrong with this line of thinking? Anthropic is the power company that has a 3D printer to make a faster Maglev than anyone. If Enterprise companies are restrictive to make your own data their only
32.
by sathish316 8mo ago
Next token prediction and Hallucination as a bug. This should be of deep concern to all Frontier labs, who think Integrity and Trust is optional when LLMs are used this way in places where it's most important.
33.
by sathish316 8mo ago
I hope I’m not talking to an AI. Can you explain how you wire up ctags to Claude code and keep them upto-date?
34.
by sathish316 8mo ago
SQL is the best exploratory interface for LLMs. But, most of Observability data like Metrics, Logs, Traces we have today are hidden in layers of semantics, custom syntax that’s hard for an agent to translate from explore or debug intent to
35.
by sathish316 8mo ago
Google is a copycat in AI products. Gemini Chat: ChatGPT Gemini CLI: Claude Code Antigravity: Cursor Nano banana: Midjourney Subscription API ban: copied Anthropic NotebookLM seems to be the only exception, or it could be an acquisition. Su
36.
by sathish316 8mo ago
Historically, Claude code used sequential planning with linear dependencies using tools like TodoWrite, TodoRead. There are open source MCP equivalents of TodoWrite. I’ve found both the open source TodoWrite and building your own TodoWrite
37.
by sathish316 8mo ago
It might also be a reflection of the number of frequent outages of GitHub under Microsoft recently and GitHub Copilot push
38.
by sathish316 8mo ago
Gemini Fast and Thinking failed just like other models. I found Gemini Pro to be more consistent in logical reasoning
39.
by sathish316 8mo ago
To build your own STT (speech-to-text) with a local model and and modify it, just ask Claude code to build it for you with this workflow. F12 -> sox for recording -> temp.wav -> faster-whisper -> pbcopy -> notify-send to know
40.
by sathish316 8mo ago
Peter clarified “I don’t read code” part in Lex Fridman interview - he said “I don’t read the boring parts of the code” that are about data transformation or writing to/reading from databases. He distinguished between what he calls “Ag
41.
by sathish316 8mo ago
I can feel the AGI on this one :) I ran extensive tests on this and variations on multiple models. Most models interpret 50 m as a short distance and struggle with spatial reasoning. Only Gemini and Grok correctly inferred that you would ne
42.
by sathish316 8mo ago
I think the blog says @steipete sold his SOUL.md for Sam Altman’s deal and let down the community. OpenClaw’s promise and power was that it could tread places security-wise that no other established enterprise company could, by not taking i
43.
by sathish316 8mo ago
Thanks for building Moltis. How does Rust alone make it more secure and immune to Prompt injection attacks?
44.
by sathish316 8mo ago
Self modifying part is through skills: Assistant -> Skill -> Script. As you keep modifying or adding capabilities, it’s the script that gets modified. Atleast that’s how OpenClaw works, I’m not sure how Moltis implements self-modifyin
45.
by sathish316 8mo ago
Don’t install skills in OpenClaw or Moltis for security reasons. Self-extension or self-evolving nature means that you can customise it to create your own skills
46.
by sathish316 8mo ago
> I haven’t been to get them to do something totally out of distribution yet from first principles. Agree with this. I’ve been trying to make LLMs come up with creative and unique word games like Wordle and Uncrossy (uncrossy.com), but s
47.
by sathish316 8mo ago
13 Markdown files have erased 285 billion dollars of wealth in the stock market - https://martinalderson.com/posts/wall-street-lost-285-billio... Why are you surprised that it takes only 65 lines of Markdown to create
48.
by sathish316 8mo ago
Read the following books: * SICP - Structure and Interpretation of Computer programs * AOA - Art of Assembly language programming * Hello Python - Think like a computer scientist * The C Programming language - Kernigham and Richie * Pick on
49.
by sathish316 8mo ago
There are several security flaws in OpenClaw: 1. Prompt injection - this is unsolvable until LLMs can differentiate command and text 2. The bot can leak secrets. The less secrets, API keys, passwords you provide the more useless it is 3. Th
50.
by sathish316 8mo ago
The lethal trifecta needs the right cocktail of foolishness to become a major security incident or scam: a millionaire or billionaire, an AI browser such as Comet or Atlas tied to personal email and banking, and any untrusted Reddit post, t
51.
by sathish316 8mo ago
I’ve come across dcg - destructive command guard - that claims to have a fast rust based runtime, with prehooks to audit any tool or command executed by an agent and to block them if they fall in some dangerous patterns - https:/&#x2F
52.
by sathish316 8mo ago
DCG or destructive command guard seems to work like the latter to prevent as many unsafe commands as possible - https://github.com/Dicklesworthstone/destructive_command_gua...
53.
by sathish316 8mo ago
How is NanoClaw immune to the Lethal trifecta attack based on prompt injection that OpenClaw is also prone to? https://simonwillison.net/2025/Jun/16/the-lethal-trifecta/ Lethal trifecta: 1. Access to you
54.
Treat OpenClaw as untrusted intern with Persistent Id and Ephemeral credentials(twitter.com)
1 points by sathish316 8mo ago|1 comments
55.
by sathish316 8mo ago
Some of the hardest problems with Remote Personal Agents from a Security perspective are: 1. Persistent id 2. Ephemeral credentials to access API keys 3. Periodic rotation and expiry of keys 4. Prompt injection and Lethal trifecta attacks f
56.
by sathish316 8mo ago
> “vibe coding has an addictive nature to it, you write some instructions, and code that looks correct is generated. Bam! Dopamine hit! If the code isn’t correct, then it’s just one prompt away from being correct” The reason Claude code
57.
by sathish316 8mo ago
Anyone who thinks they can avoid LLM Prompt injection attacks should be asked to use their email and bank accounts with AI browsers like Comet. A Reddit post with white invisible text can hijack your agent to do what an attacker wants. Even
58.
by sathish316 8mo ago
I have modified twitter thread to highlight open-source Piper TTS and non-proprietary voices from ElevenLabs or FishAudio and hidden this post now since i'm unable to edit or delete
59.
by sathish316 8mo ago
It’s only meant as a demonstration of what is possible with OpenClaw and Fish Audio. Please use open source Piper TTS or non-proprietary Eleven labs/Fish audio voices as mentioned in X thread.
60.
by sathish316 8mo ago
This is built using OpenClaw’s custom skills that are created on the fly and Fish audio. Long form X article - https://x.com/sathish316/status/2020106838419861680?s=46 It feels surreal and futuristic to get all ne

More ›