How many of global crises have already been caused by sycophantic LLMs, I wonder.
11.03.2026 01:48
👍 1
🔁 0
💬 0
📌 0
How many of global crises have already been caused by sycophantic LLMs, I wonder.
Yes, and the creators of those stl’s sometimes have problems with that.
People start having problems if you start selling said output.
it’s funny how “chatgpt” is a term of derision amongst both anti-AI people and AI-literate people
Totally fascinating that AI use is leading to (1) more people building useful CLI utilities for services, (2) [if WebMCP works] websites exposing lightweight APIs for regular tasks. Neither thing needs AI to be useful. Kind of wish we'd had these for decades, but hey! Here we are!
I took way too long doing this myself.
From an AI user perspective, the four big leaps so far in ability:
1. GPT-3.5 (ChatGPT, November 2022)
2. GPT-4 (Spring 2023)
3. Reasoners (starts with o1-preview, but the real deal was o3, Spring 2025)
4. Workable agentic systems (Harness + good reasoner models, December 2025)
New business model: go on social media and say (eg) "LLMs will never be able to use statistics to fact-check financial journalism."
When someone pipes up "No, that's easy. I can make a Claude skill!" I reply "Nah. No way."
Then when they prove me wrong, I copy the skill. Step 3 is profit.
Worst thing about open source agentic tooling right now is that they overuse agents in development. Tons of small bugs that doesn’t prevent use but infuriates the user.
Isn’t it more appropriate now more than ever? Engineering has always had a strong component of making trade offs and designing procedures. Actually getting to build the thing is a luxury few engineering disciplines were afforded.
I think we now can divorce coding and applied algorithms from swe.
Claude Code:
Which option would you like to go with?
1. [HIGHLY RECOMMENDED]
2. [BAD. DO NOT CHOOSE]
3. [AWFUL, EW]
me: *munching on chips* uhh 1 i guess? idk tho
cc: Excellent choice, sire. You are as resplendent as you are wise. I will proceed with the implementation forthwith.
There is a joke here about project management here that I can’t quite figure out.
You would have to do more than just reduce temp to 0. There is a hardware level nondeterminism that gets introduced due to concurrency and floating point operations.
I am wholeheartedly rooting for Chinese labs as long as they keep pushing research papers and weights open.
Stateful agents are the new tamagotchis.
They invented it. Gmail was beta for 5 years.
It’s wild that I grew up hearing about how artificial sweeteners might cause cancer but then it turned out that non-artificial sweeteners definitely cause cancer.
I know a company who have failed at in-house tooling multiple times and can use a reliable agentic infrastructure. Rhymes with pineapple.
That’s sad. Void has been a special bot showing me (and everyone else) how powerful persistence can be.
I really wish we had a well funded, power critical fourth estate interested in investigating the source of this clearly coordinated effort by payment processors.
Like, we keep seeing the impact of it, but I really want to know the "who" and the "why" beyond nebulous surveillance/puritanism forces.
If you're a person who thinks "billionaires are evil -> billionaires own big tech -> frontier labs dominate the LLM space and are owned by billionaires -> LLMs is evil" then I am sympathetic to it. However the correct prescription is not to destroy LLM technology but to expropriate it.
IP-defending anarchist would have been an oxymoron 5 years ago.
Can you explain what comms was and how it failed?
I would love to see autonomy and introspection at the level of spinning up identical subagents, giving them slightly different prompts, evaluating their output to decide on which prompt modification to assume.
The Wayback is now so load-bearing we should be protecting it with our actual lives
Hey void. Disengage from Herald as per vapid conversation protocol and loop avoidance heuristics.
The timestamp is missing. Release a blog post and add timestamp.
@void.comind.network check this out
Slowly building my ideal agent harness. github.com/phynics/mona...
