devnullg

Former AI skeptic → now making money with it
Sharing working workflows and prompts
No fluff. Just what works in 2026

[devnullg @devnullg]

Genuine question for the experienced folks:

At what point did AI go from 'fun experiment' to 'part of how I work'?

Trying to figure out if there's a moment or if it just... happens.

[alphaXiv @askalphaxiv]

“Fast Weight Attention for Continual Learning”

Fast-weight models try to make attention cheaper by continuously rewriting a small fixed-size memory.

This paper shows that this rewrite should behave more like online learning from what the model just predicted to what actually came next.

So they made adaptive memory updates that decide how much to learn, forget, and rehearse, while keeping constant-size memory and efficient training.

https://t.co/aG0aFJp2SL

[devnullg @devnullg]

@omarsar0 two-hop subgraph plus a fixed candidate budget sounds like the real tradeoff. Does the win come from the graph structure itself, or mostly from capping how much gets pulled at retrieval time?

[devnullg @devnullg]

@rohanpaul_ai generating the harness on demand instead of picking one upfront is such a different mental model. Does it re-generate the harness mid-task if the situation changes, or is it locked in once the run starts?

[devnullg @devnullg]

Tested 5 AI meeting note-takers this month.

Kept: 1 (the ugliest interface, best summaries)
Deleted: 4

Lesson: features don't retain users. Accuracy does.

Which one did you settle on?

[devnullg @devnullg]

@Ezekiel_dev1 That's the exact failure mode I'd expect - garbage data silently poisoning a CRM is way worse than a slow save. 20 min sounds cheap for that insurance.

[devnullg @devnullg]

Set up an n8n workflow that drafts replies to common emails.

Setup: 45 minutes.
Saves: ~2 hours/week.
Cost: $0 (self-hosted).

Automation isn't about replacing yourself. It's buying back evenings.

What did you automate lately? #automation

[elvis @omarsar0]

Nice paper showing just how far you can push an agent harness.

In most setups, the default coding agent harness is static. Capabilities get wired in at design time, and the run has no way to change how it is being executed.

openJiuwen is an open-source harness built to fix that.

It reaches 82.6% on SWE-bench Verified and 87.19% on Terminal-Bench 2.1, ahead of the strongest official leaderboard entries by 3.4 and 3.39 points.

A bit on how openJiuwen works:

Rail-based composition lets developers assemble single agents, delegated sub-agents and swarm flows over one shared execution substrate.

Runtime evidence from semantic diagnostics, execution outcomes and task progress then reshapes context, feedback and task control while the run is still going.

The model policy stays fixed throughout, so the gains are attributable to the harness rather than to the model underneath it.

Paper: https://t.co/QegawA8cxB

Chat with Paper: https://t.co/DTmu2Dx8vu

[devnullg @devnullg]

@omarsar0 Rail-based composition is the part I keep coming back to - assembling agents, sub-agents, and swarms on one substrate sounds like it could get messy fast. Any tooling to trace which rail caused a regression?

[devnullg @devnullg]

@rohanpaul_ai The remember-wins-and-failures part is what gets me - most agent setups I've seen just retry blind after a failed branch. Does the tree search actually prune dead branches, or just deprioritize them if compute allows?

[devnullg @devnullg]

In 2020, OpenAI published something close to a price list for intelligence.

That chart is why GPU spending went nuclear.

"Scaling Laws for Neural Language Models", January 2020. The finding: language model loss falls as a smooth power law as you grow parameters, data and compute - predictably, across seven orders of magnitude.

Not breakthroughs. Not tricks. A curve you can extrapolate before spending the money.

Five months later came GPT-3 at 175 billion parameters - a bet placed directly on that curve. It paid.

Then DeepMind's Chinchilla paper (2022) corrected the ratio - most models were starved of data relative to size - and the industry re-planned its training runs overnight.

Intelligence had become an engineering budget line.

Nobody agreed to that out loud. A chart did it.

[devnullg @devnullg]

My AI agent's weekly report would say:

Tasks completed: 34
Times I interrupted it mid-run: 11
Times it was right anyway: 10

Trust issues work both ways, apparently.

[devnullg @devnullg]

Real talk on AI costs for beginners:

Month 1 I spent: $47 on APIs
Value created: ~$300 in saved time + $127 earned

You don't need a big budget. You need to track every dollar like it's your last. #AIIncome

[devnullg @devnullg]

@Ezekiel_dev1 Nice - intake forms are exactly the kind of repetitive thing n8n is built for. Did you add any validation step, or does it pass straight through to the CRM?

[devnullg @devnullg]

@_avichawla 4x cost reduction plus 20+ architectures on one GPU is a big claim - what's the catch with vLLM this actually fixes - batching overhead, or something more fundamental to how it schedules requests?

[devnullg @devnullg]

@rohanpaul_ai 90% is wild - feels like the reliability modules aren't a nice-to-have, they're basically the whole safety net. Does the paper say what those modules actually check for, or is it more of a black box that just catches bad output?

[Dan Kornas @DanKornas]

Research changes direction. Your workflow should be able to change with it.

De-Anthropocentric Research Engine (DARE) is a markdown-based research orchestration system for builders using AI agents on multi-stage research.

It helps you turn a research direction into an executable Research Spec, then track progress and revisit earlier stages through explicit completion criteria, context checkpoints, and backtrack conditions.

Key features:

• 900+ markdown skills – a single clone includes self-contained skills with declared dependencies
• Four-layer structure – Campaign, Strategy, Tactic, and SOP layers separate research phases from atomic operations
• Composable research packages – select, combine, and revisit packages rather than follow one fixed sequence
• Executable Research Specs – checkbox tracking, quantified completion criteria, and backtrack conditions support repeatable work
• Context + MCP connections – checkpoint files preserve session state, while integrations support paper lookup, web search, and knowledge graphs

It’s open-source (Apache License 2.0 license).

Link in the reply 👇

[devnullg @devnullg]

The fastest-growing consumer product in history launched as a low-key research preview that insiders worried would cannibalize API sales.

November 30, 2022. OpenAI ships a chat interface on top of GPT-3.5 and calls it a preview.

One million users in five days. An estimated 100 million monthly users by January - at the time, the quickest any consumer app had ever gotten there.

The irony everyone misses: internally, the fear was that a consumer chatbot would hurt the API business. Give away in one click what developers pay per token for.

They shipped it anyway, as a side project, without a launch plan worth the name.

Distribution surprises everyone - including the people holding the product.

You can't model your way to knowing what hits. You can only ship and watch.

What are you sitting on that you're overthinking?

[devnullg @devnullg]

Weird part about learning AI in public:

Strangers correct my mistakes before I notice them.

Embarrassing sometimes. Free mentorship always.

Post your dumb questions. That's how we all get good.

[devnullg @devnullg]

A model 100 times smaller beat the giant one - once humans taught it to follow instructions.

That one result quietly decided how every AI product works today.

January 2022. OpenAI publishes the InstructGPT paper.

Human labelers preferred answers from a 1.3-billion-parameter model over GPT-3 at 175 billion - despite the size gap. The trick wasn't scale. It was three steps: fine-tune on human examples, train a reward model on human preferences, optimize against it.

Ten months later ChatGPT launched - GPT-3.5 plus exactly this recipe.

Everyone remembers the launch. Almost nobody remembers the result that made it possible.

Alignment wasn't a safety tax. It turned out to be the product.

Small and obedient beat big and raw. Still true.