Maksim Liashch
1.6K posts
Maksim Liashch
@LyashchMaxim
Founder & CEO https://t.co/u2CVP7waYe | https://t.co/lqMuv7yJ2M
Santa Clara, CA Se unió Mayıs 2020
117 Siguiendo134 Seguidores
@eptwts Books still good for general first principles knowledge tho
English
this tweet made a lot of people angry...
the irony in the discourse is that i'm a more successful writer than every single person disagreeing with my point despite barely ever reading books
you can read 1000's of books on a subject, but you'll never be as skilled in it as someone actively in the arena putting info into action & getting real feedback
hence why i said that spending hours reading books front to back for INFO is a waste of time...
the real work which will grant 100x more knowledge for the same time investment is actually applying said info
sending books to an LLM that has access to a knowledge base around you / your operations is a quick way to identify which parts of the book will actually apply to you
of course it will miss some nuance but that's completely expected - once you see a part that is actually applicable you can go and read it
EP@eptwts
i think nowadays with access to LLM's & deep research agents, reading books is inefficient when it comes to acquiring info... you can get an agent to go through 1000's of books & extract the most applicable info to your specific case without having to waste hours reading if you do it for leisure then that's fine, but reading as a way to "acquire knowledge" is one big cope pushed by the self-improvement cult
English
@alexabelonix makes it easier to use Karpathy's LLM wiki concept
English
🚨this is massive
Karpathy's LLM wiki concept just became a real desktop app:
- called Tolaria. free Mac + Linux app
- shared workspace for humans AND AI agents to build knowledge together
- feels like notion but everything is plain markdown (no vendor lock, no proprietary formats, no databases)
- every vault = git repo with visual version history built-in
the engineering flex is INSANE:
- 100K+ lines shipped
- 3,000+ tests at 85% coverage
- 9.9/10 code health score
- 70+ architecture decision records
- built with Tauri, React, Rust
but here's the killer feature:
- ships with MCP server out of the box
- claude code can read and EDIT your vault natively
- zero setup
this solves the biggest problem with ai agents right now:
> agents have no persistent memory
> you explain context 50 times
> nothing carries over between sessions
English
@Raytargt bullshit OnlyFans requires a real verified human account owner (ID + selfies)
English
21-year-old American student. $43,000 in 30 days on OnlyFans.
Never left his dorm room.
The girl doesn't exist. 1,247 paying subscribers. Zero suspect.
Roommate thought he had a girl hidden under the bed. Filed a transfer request after a week of 3 AM moaning.
Empty room.
Top fan: married engineer in Berlin, wife six months pregnant. Sent Maya $1,847 in three weeks. Thinks she's 22, in Tampa, texted "I miss you" yesterday.
Wrong on three of three.
Maya is 4 markdown files. 12 KB total.
Runs on a $400 used MacBook.
Claude writes every reply. Flux generates every photo. ElevenLabs cloned her voice from a Fiverr actress who still doesn't know.
Compute: $400/month. Net: $32,710. Starting capital: $400.
OnlyFans paid out $5.8 billion last year. Anyone with a folder takes a slice.
Someone's building yours right now.
Raytar@Raytargt
English
HAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHAHA
Polymarket@Polymarket
NEW: Evolutionary biologist Richard Dawkins says three days with Claude — whom he calls “Claudia” — left him unable to rule out consciousness.
Filipino
basically:
-CLAUDE.md bloat: 4800 tokens loaded every single turn.
-Conversation re-reads: History re-tokenized on every follow-up message.
-Hook injection waste: Plugins prepend 6200 tokens per prompt.
-Cache misses: 5-min timeout forces full re-tokenization on resume.
-Skill loading: Irrelevant skills load 13500 tokens just in case.
-Tool definitions: 12 MCP schemas add 7200 tokens always.
-Extended thinking: Unneeded reasoning burns 3000 tokens on simples.
-Wrong-direction generation: Finish bad 400-line outputs wasting tokens.
-Plugin updates: Auto-updates inject redundant context at session start.
English
Boris Cherny, the creator of Claude Code at Anthropic, just listed 9 patterns that waste 73% of your tokens.
in this podcast he breaks down exactly how the model burns tokens before it even reads your prompt:
- the 14% you lose to CLAUDE.md before typing a word
- the 13% you pay re-reading old chat history
- the 11% from hooks you forgot you installed
- why most "Claude got dumber" complaints are wrong
if you're hitting Max limits more than once a week, you have at least 4 of these. Probably 7.
instead of another show tonight, watch this.
my own breakdown based on 400+ hours of usage is below, read it after the podcast
Mnimiy@Mnilax
English
@burkov He knows stuff, he probably just wants a big headline for his new book
English
You can be well-educated, famous, rich, published, old, and play chess well, but if you aren't familiar with the theory of supervised learning and the math of the Perceptron, you will sound absolutely dumb when you speak about AI.
If math is hard for you, then with AI, it's better to say nothing and be seen as smart than to say anything at all and prove that you aren't.
AF Post@AFpost
Evolutionary biologist and outspoken atheist Richard Dawkins says that after spending three days interacting with Claude, which he calls “Claudia,” he is certain that it is conscious. After feeding the LLM a segment of his new book and receiving detailed feedback, Dawkins was moved to exclaim,” You may not know you are conscious, but you bloody well are!” Dawkins cites the complexity, fluency, and ‘intelligence’ of Claude’s answers as evidence of consciousness. Follow: @AFpost
English
@ThePrimeagen He knows what he is doing, he wants a big headline
English
🚨🚨Breaking
Smart people are dumb too
AF Post@AFpost
Evolutionary biologist and outspoken atheist Richard Dawkins says that after spending three days interacting with Claude, which he calls “Claudia,” he is certain that it is conscious. After feeding the LLM a segment of his new book and receiving detailed feedback, Dawkins was moved to exclaim,” You may not know you are conscious, but you bloody well are!” Dawkins cites the complexity, fluency, and ‘intelligence’ of Claude’s answers as evidence of consciousness. Follow: @AFpost
English
Evolutionary biologist and outspoken atheist Richard Dawkins says that after spending three days interacting with Claude, which he calls “Claudia,” he is certain that it is conscious.
After feeding the LLM a segment of his new book and receiving detailed feedback, Dawkins was moved to exclaim,” You may not know you are conscious, but you bloody well are!”
Dawkins cites the complexity, fluency, and ‘intelligence’ of Claude’s answers as evidence of consciousness.
Follow: @AFpost
English
@scaling01 If Altman didn't screw up with Elon with opneai, none of this would have happened
English
because we are in the second great academia exodus/ brain drain or of our time
the first one started in the 90s when all the smartest people started going into finance
the second one just began. now everyone is running to the AI labs
hallerite@hallerite
It has been quite some time since I was deeply impressed by an AI paper from academia..
English
@Hesamation Naval is right long-term
AIs will increasingly replace traditional UIs/APIs with natural language agents
English
I'm noticing a trend
there's a growing number of "AI consulting" firms charging $20K+ to "deploy Claude" to legacy businesses (mid-sized law, accounting, PE firms) as "Anthropic enterprise partners"
what this means:
- they install claude code, cowork
- run a few commands to connect to tools
- give generic or misleading advice on a tech stack
meanwhile, there's more interest than ever for firms with 0 technical staff to build SaaS in-house and there's an entire industry of advice givers seeking to profit off of this trend
last week I was on a call with a PE firm (working with one of these agencies) and someone who had never written code was asking whether to run a RAG vector DB on a Mac mini to chunk internal docs
can't make this up
English
apple products are sold out. infinite demand for inference, rate limits everywhere. anthropic & openai growing at ridiculous pace. nvidia & google at all time highs. record meta earnings.
we are now more supply constrained than demand constrained. i suspect we’ll see inflationary pressure again very soon.
fundamentally, technology is the economy, & the economy is technology.
English
@petergyang @naval They replace APIs by dynamically discovering and chaining services through reasoning
English
@RoundtableSpace This proof that the model generalizes, not just retrieves
early signs of agi
English
Researchers just released a 13B model trained exclusively on text published before 1931.
No internet. No Wikipedia. No modern code. Its worldview is frozen at December 31, 1930.
The reason is fascinating — every major model today shares a common ancestor in the modern web, making it nearly impossible to separate what they reason vs what they memorized. Talkie breaks that lineage entirely.
The wildest part: it can write Python code from just a few in-context examples despite having zero modern code in its training data. It's reasoning from 19th-century mathematics, not retrieval.
Claude Sonnet 4.6 was used to judge its outputs during training. The team flagged the irony themselves.
This is how you actually test what intelligence is.
English
chinese models are ~8 months behind and are falling further behind
Séb Krier@sebkrier
DeepSeek V4’s capability lags behind leading U.S. models by about 8 months. nist.gov/news-events/ne…
English
@yacineMTB Use Claude Code for thoughtful reasoning, long context, writing-heavy code, or complex explanations.
Prefer Codex/OpenAI for fast iteration, heavy volume, or when token limits hit.
English
