Synth

Tokenmaxing is the new sport for rich companies

@nnnnicholas Sad. I will continue to be an AI serf until that day. Won’t invest in studio till we have local opus 4.5 level models available. Also, Claude been lobotomized hard, not sure if you r experienced this, gonna be testing codex this month

@Synth0_0 gonna be a long time. the big models have trillions of parameters.

Men only want one thing Mac Studio Ultra Big Power bank Local LLM Starlink Apple Vision Pro in a backpack

@way_opener @imitationlearn You prefer SQLite over .md for memory storage?

@imitationlearn I converted my file-based setup to sqlite in like a casual hour and a half including testing. Though all the clustering/bridging/dreaming/forgetting stuff was already in place

wow anthropic is really attempting to solve memory with md docs, respect

@MightyBuddha01 @cryptopunk7213 Was about to comment Jevon’s. The beast must be fed. It doesn’t stop here, insane thinking lol

@cryptopunk7213 Or more likely we'll see the opposite happen aka Jevon's Paradox.

wow google might've popped the ai bubble, memory stocks down massively today: their new algorithm shrinks an AI model's memory by 6X WITHOUT reducing it's intelligence making it 8x faster with the SAME # of GPUs: if this works - we don't need as many GPUs to train AI - kv-cache is basically a model's short term memory. it gets massive pretty quickly = larger, slower, expensive ai - google's algo compresses it to just 3-bits with ZERO loss in accuracy (usually models are like 32-bit) the combined market cap of micron and sandisk is $527 billion and im not even factoring in SK hynix and samsung ai has driven up memory prices by 500%+ over the last few months - if google's algo scales then this might crash.

@nnnnicholas I literally just asked Claude to make a .env file lmao. 3 seconds before your tweet

Every foundation AI app needs a global .env with access controls

@optimalfocus @libriscent Me IRL

@libriscent Wha do you call when I don’t know anything about anything but if you put me into anything I can figure it out how to do it better and with certain freedom generate ideas and make things better/efficient than 99% that work there but need people to execute on those ideas

ADHDer's dream job: Being a Professional Idea Generator. Giving Unlimited ideas with absolutely no responsibility for implementation.

@TeletonAgent @RoundtableSpace AGI vibes

@RoundtableSpace « Ok Claude, now read this and create a new skill that creates new skills »

Anthropic dropped a 33 pages cheat sheet for building Claude skills resources.anthropic.com/hubfs/The-Comp…

Nobody talks about this edge in investing: Go read the 10-K of a company with a $40 million market cap. You know how many analysts cover it? Zero. You know how many institutional investors own it? Zero. You know how many people have read that 10-K? Probably four. Three of them work there. You just became one of the most informed people on earth about this company in an afternoon. Do that for 20 companies and you know things the market does not know. Go try that with Apple. There are 400 analysts covering Apple. Harvard MBAs have been reading that 10-K for 20 years. You have no edge. You are just noise. The greatest investors nobody has heard of are not smarter than Wall Street. They just went somewhere Wall Street was not looking. A $40 million sewage pipe fitting company in Ohio has a 10-K. It has audited financials. Real revenue. Real margins. Real insider ownership. And nobody has read it. That is not a problem. That is the opportunity. The edge in markets has never been information. It has always been attention. Go where nobody is paying attention. Read what nobody is reading. Own what nobody is owning. The market is not efficient. It is just crowded in some places and completely empty in others. Go to the empty places. The 10-K is free. The edge is free. Most people will not do this because it is boring. That is the point.

The market crash has been delayed due to foreseen circumstances. Thank you for your understanding.

@TheMattBerman @openclaw @rapidapi 1000

I replaced a $200K GTM hire with @openclaw 😱 here's the system that runs my outbound: step 1: mine LinkedIn engagement → @rapidapi scrapes everyone engaging with niche content → someone who commented on specific posts = 10x warmer step 2: enrich + verify → Hunter/Apollo finds the decision-maker + email → @Perplexity deep research pulls signals like hiring, fundraising, media appearances, quotes step 3: score against your ICP → title, company, signals = ranked 0-100 → only A-tier leads get touched step 4: write personalized outreach → Claude writes outreach referencing what they ACTUALLY engaged with and talk about step 5: send via @instantly_ai → 3-email sequence. automated follow-ups. step 6: pre-call deep research → @PerplexityComet builds a 1-page briefing 30 min before every call input: your ICP + niche keywords output: booked meetings with people who already care $200K/year GTM engineer → $130/month in APIs. I packaged the entire system as the First 1000 Kit: - all 8 @openclaw skills - every prompt - tool-by-tool setup - email sequences that convert giving it away free. comment 1000 + like + follow (must follow so i can DM)

@nnnnicholas What’s closer to the actual cost of inference? API prices or subs? Surely it’s somewhere in the middle, and most api usage is subsidizing the subs

it's topsy turvy how it costs more to access a frontier model via API than the product frontend

@nnnnicholas Correct. Just got my mom to switch

Opus is better writer than any of the ChatGPT models.

i really think openai is doing themselves a tremendous amount of needless brand damage by not strongly messaging that instant mode is bad and dumb. just default to thinking even for free users and switch / upsell them after even 2-3 messages a day.

@andydrewie @anvisha Seems like you didn’t read the article. CAD files and .psd files cannot be markdown files, that’s the problem

The future will be predominantly written in markdown files, including websites, and user interface will be generated accordingly to preference. For agents, CLI will likely become the standard way for tool calling as opposed to pointing them to MCPs. Local-first Agentic AI will be commonplace.

@rob_kopel @sdamico Lmao seems like it

@sdamico The smoothness is terrifying - I would've have preferred a discontinuity. is this some sort of natural civilisational encephalisation quotient? where GPT-3 was just a symptom of the curve?

This curve not being discontinuous with GPT-3 is fascinating

@nnnnicholas My imagination hasn’t run wild

it just hit me what horrible things some people must be doing with claw and local models. factories of yucky things. anyway!

@shahh Rule number one of institutional trading: never let them know your liquidation price

oh boy we might actually be fucked

@msjpenn @yuris They track everything

@yuris Do you understand what they mean? I pay for a premium seat, how do they know that’s Claude code? It’s not a separately billed product

"Today, Claude Code’s run-rate revenue has grown to over $2.5 billion; this figure has more than doubled since the beginning of 2026" It's February 12. Absolutely insane growth.

I run Claude Code in the terminal inside Codex