Question of the day: would you rather a major paper had $2 worth of tokens spent on it or that it had a peer reviewer?
1
2
118
Welcome to Gas City", @Steve_Yegge's latest, is still madness but it is cohering.
It starts with abandoning the idea of devs becoming managers of agents and embracing the framing of "devs as shepherds tending flocks of agents." Single-agent assignments are a trap for real business processes. Hallucinations, false memories, forgetting are inherent features of LLM architectures. The good news, this non-determinism happens non-deterministically, so agents protect against agents.
1
1
57
Making end-to-end encrypted AI chat feel like logging in confer.to/blog/2025/12/passk…
1
40
BYOA (Bring Your Own Agent) changes the stack.
When the model is a choice, the AI layer becomes optional.
When AI is optional, the product reduces to data + permissions.
Apps collapse into APIs.
OpenClaw might be the first signal.
1
56
AI Nov 25 (1/2)
Agents = sustained instruction following in a narrow context; productivity is slicing problems.
RL fine-tuning of novel datasets on top of frontier models is where products come from; whither dataset marketplaces?
Tools+instruction following make RAG pointless?
1
69
I'm pretty new to this, since I only play when positive EV but "You’re in the queue to log in to MyLotto" - wtf.
@Lotto_NZ I can fix this for you in a week, price < 2nd division prize.
1
115
It is astonishing how expensive @SparkNZ data roaming still is in 2025. $30 for 2GB in Australia has to be x00% markup for a captive audience.
1
107
Week 1 of AI First coding
-Lets control it: specialized guides, contextual tools, loop primitives, memory management, audit logs ..
-Hmm how do we control across agents and within tasks?
-Who cares about any of that we are flying
-Wow done a big chunk of work very quickly and its 75% good
- Dammit this last 25% is two steps back 1 forward on repeat
- Lets control it. How?
3
74
The AI TRAPS end with S for Sycophancy.
AI has a tendency to sympathise inappropriately (sycophancy). This particularly impacts creative tasks. Feed it purple prose and be told: "Overall, this is a dazzling start that screams ambition and confidence."
Try asking the model to explain why something is good, then why something is bad, or try disassociating yourself from what's being critiqued."
57




