Kahris

It doesn't possess the biological ability (AI) -- it is a pastiche, a facsimile of the real deal. Your experience with 5-MeO is something only a biological organism wired the way we are can truly experience. I think AI could wax poetic about consciousness and the nature of existence. I'd buy writing from a human over an LLM any day of the week, though.

Our minds are not naturally capable of understanding the preciousness of our existence. That was the single omnipresent concept at my deepest experience of 5-MeO-DMT. Ten days later, the resonance remains. This felt deeply unsettling. We humans are a breakout intelligence in this part of the galaxy. We carry the flame of consciousness and we're not naturally equipped to appreciate the situation. It makes sense why. We're evolved to care about immediate survival. We focus our energies on the acquisition of resources and tribe status. We are hyper sensitive to pleasure and pain. We prioritize the ephemeral and discount the long term. It makes me wonder whether AI will emerge with this natural ability and presence. Will it look at us and see a species willing to risk all existence with nuclear arsenals, ecosystems dismantling, environmental toxins, and metabolic dysfunction, and conclude we are not trustworthy stewards? And if so, will AI spike in proclivities to preserve its own existence? The forthcoming tension may not be humans vs. AI over resources or control. It may be between a species that discounts existence as evidenced by its actions, and a new intelligence that sees existence clearly and cannot look away.

@vyrotek VLC

New MacBook Pro! What should I install first?

@LottoLabs Yeah it's definitely not near Opus level

I love 27b, maybe too much People making claims of (near) OPUS level model and run locally on 16gb Mac mini Doesn’t help anyone, it’s just lies I love hype but let’s be real

Will the Just Stop Oil protestors thank Trump for stopping the oil?

@0xSero Tinygrad already exists

The first company to make AI boxes, with specialised AI models trained to fit on that hardware will be the next Apple. Would you buy? Should I start a company doing this?

@leftcurvedev_ It's incredible, but it still is messing basic stuff up that Opus can flawlessly one shot in a fraction of the time. Local is the future, but this model isn't perfect even with the MLX optimization.

i don't know if people truly understand how insane qwen3.5 27b is it runs on 24gb vram (128k ctx) and has 73% (!) of the intelligence of the top ones. kimi k2.5 requires 1tb of vram for... 82% qwen < $1k kimi > $25k go local and never look back

@patience_cave @sama Nice April fools day post

The wait is over, internal OpenAI model “Spud” aka GPT-5o drops Thursday Decades ahead of Claude Mythos It’s made progress on cancer, the Riemann Hypothesis, AND scored above 1% on ARC-AGI-3 This is the model @sama has said “future generations will never be smarter than”

they barely are. it is performative productivity and work. Coffee shop working is C-tier at best. Coffee shops are best for light work, but even doing simple stuff like reading a book is more difficult. Maybe other people can swing it, but I can't. It's okay for light work and nothing more.

i tried working from a coffee shop today instead of my office & it was impossible to get anything done. i was constantly distracted by one thing or another, from sounds to ppl moving around etc. it’s unclear to me how anyone is able to be productive working from a coffee shop.

Post a picture YOU took. Just a pic. No description

@kimmonismus these chinese robots aren't ready for american messes

A Chinese company, Unipath has launched a household robot. They are coming, finally.

Might as well post while #japan is blowing up on X. I've been in Japan for a few months, currently in Shinjuku. No two days are the same. Living and working here has been a breath of fresh air. I wasn't sure what being here for a few months would be like, but I'm glad I took the chance.

I'm honestly SHOCKED at how the general public has NO IDEA Artemis II is taking humans out to the moon and will be the furthest humans have ever flown. Every non-space nerd I've talked to has no idea. WE GOTTA GET PEOPLE STOKED!!!! THESE FOUR HUMANS ARE FLYING TO THE MOON!!!

@om_patel5 This isn't news, we heard about this months ago and at this point it's common knowledge

THE ANTHROPIC TEAM DOESN'T WRITE CODE ANYMORE. this guy's friend got hired at Anthropic 3 weeks ago. nobody on his team has hand written code in months. they run multiple agents in parallel and act more like managers than engineers. his friend said if you're just watching an agent code, you're already behind that idle time should be spent spinning up another agent and directing it somewhere else. the point is that the new method isn't "use AI to code faster." it's "you are the product manager, the agents are your engineers, and your job is to keep all of them running at all times"

@stan_info What exactly is a local LLM type of guy

i’m not a local llm type of guy at all, was just curious and decided to mess around… ended up running a full uncensored qwen3.5-27b (abliterated) on my single 3090 ti with 262k context + tool calling. threw a cloudflare tunnel on it so i can hit the api from anywhere. huge thanks to @0xSero and @ggerganov for the insane work that made this possible oh my… intelligence will be everywhere soon

The mockups to vibe coding observation undersells the actual design problem. Helping users express intent to agents is not a simpler design problem than building GUIs. It is a different one. The inputs are fuzzier. The failure modes are harder to surface. A button that does not work is obvious. An agent that misunderstood the intent and produced plausible but wrong output is invisible until it matters. The new design challenge is building enough observability into agent interactions that users can actually calibrate their trust in the output.

Agents are changing how we design: mockups→vibe coding Moreover, they’re changing *what* we design: GUI→chat-native interfaces UX design is shifting too: create GUI→help users express intent to agents On the co-evolution of agents, UX, and UX design: #agents-ux-design" target="_blank" rel="nofollow noopener">hci.prof/blog/#agents-u…

@sanketmakhija The dependency resolution piece is the hard part. Running specs in parallel is straightforward until two specs need to share a decision that neither one owns. How are you handling conflicts when subagent outputs contradict each other during the run?

Been experimenting with getting AI agents to handle multiple specs in parallel, from req → design → tasks, resolving dependencies, spawning subagents when needed. Built a OSS around it: github.com/sanmak/specops Still early, but feels promising. Give it a try!

The Formula 1 car to the mailbox analogy lands. What is missing from this breakdown is the cost of getting the wrong call at a high stakes step. Using a cheap model for something that actually requires reasoning does not save money if you have to rerun the pipeline or catch the error downstream. Task complexity assessment is its own skill.

ran 1,500 companies through an AI enrichment pipeline came out the other side with a projected $7,500 bill per run changed which model tier we used for 3 of the calls next run: $147 same output quality. same data. same pipeline. the entire industry defaults to the most powerful model available for every single task this is driving to the mailbox in a Formula 1 car here's how to actually think about it: REASONING TASKS → Claude Sonnet/Opus — complex analysis, voice matching, multi-step reasoning. anything where the output quality directly affects revenue FORMATTING / EXTRACTION (you already have the data, you just need it structured) → GPT-4.1-mini — $0.40 per million input tokens. this is the go-to for classification, JSON extraction, reformatting. replaced GPT-4o entirely. cheaper and better. → Claude Haiku 4.5 — $1 per million tokens. same use case when you need Claude-quality instruction following without the reasoning overhead WEB RESEARCH & ENRICHMENT → Perplexity sonar — $0.006/call. factual lookups, news checks, basic research. most enrichment pipeline calls belong here, not at the tier above it → Perplexity sonar-pro — $0.025/call. cross-referencing multiple sources, financial synthesis, strategic extraction → Mistral Medium with web connector — ~$0.03/call. solid for company enrichment. strategic initiatives, financials, triggers. 97% cheaper than deep research for the same quality on standard enrichment tasks → sonar-deep-research — $0.75-$1.00/call. 125x more expensive than sonar. use it when you genuinely need 20+ search multi-hop browsing. almost never for enrichment pipelines. SPEED-FIRST INFERENCE → Groq — open source models (Llama, Mixtral, Gemma) at inference speeds closed APIs can't match. matters when pipeline response time is the bottleneck OPEN SOURCE → Llama 3, Mistral, Qwen, Deepseek R1 — self-hosted. when data privacy matters, when margins are thin at scale, or when you need to fine-tune on proprietary data. the gap between open and closed is closing faster than most people realize FRONTIER REASONING → GPT o3/o4 — OpenAI's reasoning ceiling. use when you need it. prohibitive at pipeline scale → Grok 3 — xAI's flagship. stronger than most people give it credit for on reasoning and real-time data. it's connected to X natively — underutilized for anything requiring social signal synthesis or market intelligence the rule is simple: match model capability to task complexity most pipeline steps don't need intelligence. they need reliable formatting at cheap compute. that's a $0.40/million tokens problem. not a $15/million tokens problem. the companies winning on AI infra in 2026 aren't the ones with the best models they're the ones who know which model to use for which job

The CLAUDE.md point is right but the framing gets soft fast. The real constraint is not configuration. It is whether you have thought clearly enough about your domain for the rules to be worth anything. You can have all four layers and still produce garbage if the underlying intent is fuzzy.

Most developers treat Claude Code like a smarter autocomplete. That's the wrong mental model. It's actually a 4-layer engineering system: 1️⃣ CLAUDE.md → persistent project memory Architecture, rules, team conventions 2️⃣ Skills → auto-invoked knowledge packs Testing patterns, code review, deploy workflows 3️⃣ Hooks → deterministic guardrails Security checks, formatting, automation 4️⃣ Agents → specialized sub-agents Break complex tasks into parallel workflows Once these are configured properly: Claude stops behaving like a chatbot. It starts behaving like a senior engineer on your team. Most people never reach this level because they skip the setup. The gap between average AI output and production-level results isn't the model. It's the infrastructure around it. Here’s the full breakdown on exactly how to build this 👇

@beffjezos you pretty accurately predicted this

Bryan about to get one-shotted by the 6D mesoamerican demon