permaXag

This week, I spent 8 hours reading 50+ articles about AI. And I learned more in 1 day than most will in an entire year. Highly recommend you read these 8:

10 repos that cut your ai agent token bill by up to 80% 1. microsoft/LLMLingua → cuts prompt size by up to 95% compresses prompts before the api call. 20x compression. published at EMNLP + ACL. near-zero quality loss. 6,100 stars github.com/microsoft/LLML… 2. mem0ai/mem0 → replaces full conversation history in context stores what matters. retrieves only what's needed. 10,000 token history → 200 token memory. per agent. 54,800 stars github.com/mem0ai/mem0 3. BerriAI/litellm → routes each call to the cheapest model simple task → haiku. complex task → sonnet. tracks cost per agent, per call, per day. 45,700 stars github.com/BerriAI/litellm 4. run-llama/llama_index → replaces sending full documents rag: 100-page doc → 3 relevant chunks → same answer. 98% fewer tokens per query. 49,100 stars github.com/run-llama/llam… 5. chroma-core/chroma → replaces keyword search in full context vector store. finds the closest match. feeds only that. 50-200 tokens per query instead of thousands. 27,800 stars github.com/chroma-core/ch… 6. letta-ai/letta → replaces infinite context window crashes paged memory for agents. loads only relevant memory. stops your agent from hitting limits and retrying. 22,400 stars github.com/letta-ai/letta 7. guidance-ai/guidance → cuts output token bloat by 30-50% structured generation. constrains model output natively. no more 100-token prompts to get json back. 21,400 stars github.com/guidance-ai/gu… 8. Aider-AI/aider → replaces pasting entire codebases builds a repo map. sends only files relevant to the task. not your whole project. just what the agent needs. 44,300 stars github.com/Aider-AI/aider 9. openai/tiktoken → count tokens before you send know the exact cost before the api call happens. not after the bill arrives. 18,100 stars github.com/openai/tiktoken 10. simonw/ttok → hard cap on what gets sent cli tool: count tokens, truncate to budget limit. pipe any text in. get truncated output back. 389 stars github.com/simonw/ttok most agents are expensive not because the model is expensive. because nobody checked what was being sent to it.

Did a very different format with @reinerpope – a blackboard lecture where he walks through how frontier LLMs are trained and served. It's shocking how much you can deduce about what the labs are doing from a handful of equations, public API prices, and some chalk. It’s a bit technical, but I encourage you to hang in there - it’s really worth it. There are less than a handful of people who understand the full stack of AI, from chip design to model architecture, as well as Reiner. It was a real delight to learn from him. Recommend watching this one on YouTube so you can see the chalkboard. 0:00:00 – How batch size affects token cost and speed 0:31:59 – How MoE models are laid out across GPU racks 0:47:02 – How pipeline parallelism spreads model layers across racks 1:03:27 – Why Ilya said, “As we now know, pipelining is not wise.” 1:18:49 – Because of RL, models may be 100x over-trained beyond Chinchilla-optimal 1:32:52 – Deducing long context memory costs from API pricing 2:03:52 – Convergent evolution between neural nets and cryptography

@brim006 Screw the Ash Borer

Grahams Island State Park in North Dakota. One of the few remaining healthy Green Ash Tree forests in the upper Midwest. Thanks to the bitterly cold winters, Emerald Ash Borer has mostly halted around SW Minnesota and NE South Dakota.

Karpathy didn't make a course. He made THE course. 3 hours. Free. Tokenization. Attention. Hallucinations. Tool use. RLHF. DeepSeek. AlphaGo. Every behavior you've ever wondered about in an LLM - where it comes from, why it exists, how it was engineered. The gap between engineers who understand this and engineers who don't isn't technical depth. It's the ability to conceive of entirely different things.

@Tusslindo Cost of living isn’t coming down… ever 😂😂

If you didn't have a kid by now, you just gotta hold off until the cost of living comes back down to earth.

@Say_JB_247 @ContrarianCurse If you someone was a single handicapper before and play infrequently now it’s definitely possible

@ContrarianCurse Unless you are a 1% top-tier gifted athlete or playing putt-putt, I don't believe that you shoot a 87-95 and play 4-5 times a year. Then on top of that...throwing in the astonishment for those that shoot 130 for the down low humble brag. Nothing personal, it's just not true.

I am a solid 4-5 rounds of golf/year kind of guy. I really don't love it. I'll shoot 87-95 depending on the day with always a chance of a blow up. I find it frustrating generally Never ceases to amaze me how obsessed someone can be for years while shooting like 130

@pronounced_kyle Yes. 15 “no go” 485 “go”

Many people replying to this clearly didn't learn the word "plurality" in college

If you want to be a S&P 500 CEO, the best college to go to is "no college"

@hvsperus @DJBranham It’s not about flying to Philly. It’s getting to State College. Philly and Newark airports are both 3.5-4h drive away

@DJBranham am i missing the joke?

Imagine Europeans trying to find how to get to State College, PA

This video literally breaks down how I built a $20M/yr marketing agency and removed myself from the day-to-day:

Some of you asked to see a real run of the Council of High Intelligence — so I recorded one. Task: decide what to do with LACP next. When the council converges too early, it gets challenged. When it tries to exit unfinished, StopHook (LACP) forces completion.

@nathan_covey Ventusky is the best

I just found the greatest weather website of all time. You see, I've been scammed by the weather many times in my life. I thought Utah was hot, then I lived in Argentina and Georgia, which on paper have similar highs, only to find that 90°F in these places is 10x worse. Obviously, part of the solution is to look at the "feels-like" temperature, but I've never seen weather apps show historical or forecasted values for this. They only show today. So if you're trying to assess the heat of a place in its various seasons, whether to travel there or move there, you're out of luck. Very slim data. BUT, last night at 1 am, I finally found it: Ventusky. It shows historical data for the feels-like temperature on a great visual map. Its free version also has 7-day forecasts and no ads (which most weather apps are bombarded with). I used to pray for times like this.

If you’re looking for a business opportunity that has a high chance of success in 2026… USING AI. But you don’t want to: • Build an audience • Be the expert/face of the biz • Want a CLEAN business • Almost zero overhead… Read this:

USE THESE GITHUB REPOS TO UPGRADE YOUR POLYMARKET TRADING GAME * PMXT github.com/pmxt-dev/pmxt * POLYMARKET AGENTS github.com/Polymarket/age… * FAST MCP github.com/PrefectHQ/fast… * MCP github.com/caiovicentino/… * POLYREC github.com/txbabaxyz/poly…

@J_pxlr @AzFlin Real

@AzFlin Sonnet. Please fix this. It's fixed. No it's not. Woops sorry, now it is. No it's not. Look at this. Oh you found the problem! I have implemented the changes and you are good to go. NOT WORKING. Oh sorry about tha- *Thinking cancelled* Switch to Opus you idiot.

> opus 4.6 > thinking mode set to high > plan mode perma on > 15 diff skill MDs installed on coding best practices "yo guys wtf my claude usage already is already full"

@calebcanales_ Gotta love Sunny Isles

OH MY GOD IT DOES

@boringlocalseo Localrank

REPEAT AFTER ME LLM citations have never been more important Took 2 weeks to get our client recommended by Gemini + ChatGPT This method gets you featured in 500 listicles for less than $0.5 EACH Comment LOCALRANK + like this post, and I'll DM it to you (must be following)

@iamliamsheridan System

we analyzed 4.7M cold emails across 10+ clients. here's what actually moves pipeline (and what quietly kills it): 1/ more follow-ups hurt you. after a certain point, they don't increase replies. they kill them. most teams don't know where that point is. 2/ same email. 4.3x different results. one small change. same copy, same offer, same list. nearly 5x the outcome. 3/ the metric you're tracking looks great on paper. it's also draining your pipeline. you probably celebrate it. you shouldn't. 4/ there's a setup step almost everyone skips. not exciting. nobody talks about it. but skipping it destroys your deliverability before you send a single email. 5/ winning campaigns don't win because of copy or offer. it's something simpler. something most people walk right past. 6/ outbound isn't broken. your outbound is broken. there's a specific reason it's not working. it's fixable. comment SYSTEM and i'll send you the full breakdown.