CryptoWifΞ 🦇🔊

@karpathy When I hit the context limit and had to start a new channel, GPT offered to formulate a ‘summary’ for the new channel to echo all of the references and progress from the prev channel. I regularly purge/manage the memories so I also get GPT to summarize/consolidate a TLDR memory

@karpathy Yes! I have different channels in GPT4o (ie Book Club, Food, AI, Mental Health/Therapy) You are right that this is 100% like a moat, esp for my 'therapy' channel cuz I dont wanna start over with a new 'therapist' 😂

When working with LLMs I am used to starting "New Conversation" for each request. But there is also the polar opposite approach of keeping one giant conversation going forever. The standard approach can still choose to use a Memory tool to write things down in between conversations (e.g. ChatGPT does so), so the "One Thread" approach can be seen as the extreme special case of using memory always and for everything. The other day I've come across someone saying that their conversation with Grok (which was free to them at the time) has now grown way too long for them to switch to ChatGPT. i.e. it functions like a moat hah. LLMs are rapidly growing in the allowed maximum context length *in principle*, and it's clear that this might allow the LLM to have a lot more context and knowledge of you, but there are some caveats. Few of the major ones as an example: - Speed. A giant context window will cost more compute and will be slower. - Ability. Just because you can feed in all those tokens doesn't mean that they can also be manipulated effectively by the LLM's attention and its in-context-learning mechanism for problem solving (the simplest demonstration is the "needle in the haystack" eval). - Signal to noise. Too many tokens fighting for attention may *decrease* performance due to being too "distracting", diffusing attention too broadly and decreasing a signal to noise ratio in the features. - Data; i.e. train - test data mismatch. Most of the training data in the finetuning conversation is likely ~short. Indeed, a large fraction of it in academic datasets is often single-turn (one single question -> answer). One giant conversation forces the LLM into a new data distribution it hasn't seen that much of during training. This is in large part because... - Data labeling. Keep in mind that LLMs still primarily and quite fundamentally rely on human supervision. A human labeler (or an engineer) can understand a short conversation and write optimal responses or rank them, or inspect whether an LLM judge is getting things right. But things grind to a halt with giant conversations. Who is supposed to write or inspect an alleged "optimal response" for a conversation of a few hundred thousand tokens? Certainly, it's not clear if an LLM should have a "New Conversation" button at all in the long run. It feels a bit like an internal implementation detail that is surfaced to the user for developer convenience and for the time being. And that the right solution is a very well-implemented memory feature, along the lines of active, agentic context management. Something I haven't really seen at all so far. Anyway curious to poll if people have tried One Thread and what the word is.

@TrustlessState He's not training for nothing 💪

I think this man should lead the future of Ethereum

@songadaymann Officially #2 fav Song a Day (right after ETH ETF, which is still on my running list 👨‍🎤❤️)

so many great things built on ethereum, too many to fit into one song. bid on the 1/1: songaday.world/5881

Haters: The merge will never happen Core devs:

@Degen_Alfie @cobie @coinfessions @inversebrah This is stunning! Beautiful illustration

@cobie @coinfessions @inversebrah Makes it all back revenge trading. Steve Jobs-esque comeback story in the making 🎲

I have spent countless amounts of money on screens, tools, data feeds, sierracharts, even paid for a Bloomberg terminal and still haven’t been able to reliably make a profit.

GM! Saturday morning getting greenpilled with @owocki and @GitcoinDAO 😎 📖🐛

@sauceror__ Hooray! Excited to see this come to life 👏🥳

@Crypt0xWife was kind enough to mock this up for me back during the initial pengu mania. I always had visions of pudgy toys and plushies but lacked the capacity to make it a reality. I am beyond excited to see Luca & co. bringing Pudgy Toys to life.

@TheFutureisDAO1 @sassal0x This is amazing!

Im happy to say that @sassal0x has accepted my invitation to join the chippi fam! Welcome! Can you spot him walking around in Castle Ethereum?

@sassal0x @thedailygwei This one made me go for a run after a month+ slump, and now I've gone a few days in a row. Thanks bud 🤝

The latest @thedailygwei Refuel is ready for your consumption! ⛽️ Today's topics: - Market meltdown 📉 - AllCoreDevs updates 🐼 - Over 400,000 Beacon Chain validators 🥩 - and much more ➕ Watch 👇 youtu.be/A9w3w8GOp6c

@sassal0x @ASvanevik @econoar Thanks for reminding me that I don't own enough ETH

@ASvanevik @econoar Take 2mins to read this thedailygwei.substack.com/p/you-dont-own…

Prince Louis is every dev being asked for the 450,000th time

A @BoysClubCrypto Take Do Kwon is the Tristan Thompson of crypto. Remarkably consistent every time👖🔥

@sassal0x Are we adding 🐺 to the CT emojipedia? 🐂🐻🐼🐬🦇

Inside me there are 2 wolves 1 that is hyper-giga-bullish on ETH and can't stop buying it whenever possible 1 that is hyper-giga-bearish on 99% of tokens/coins and wonders how anyone can be redacted enough to buy them

@lay2000lbs We are with you @lay2000lbs! Thank you for fighting the good fight ✊🇫🇷

As some of you know, I’m currently being sued (personally) in a putative class action lawsuit. I can’t speak about it publicly because the litigation is ongoing but I want to share some context and ask for your help.

@sassal0x @tztokchad @Tetranode @ProtocolGuild 👀😔

Giga chads @tztokchad and @Tetranode have each donated 1 ETH to Ethereum public goods via the @ProtocolGuild I will be doing the dance to save the markets soon, this is not a rug, I will actually dance - but I want moar ETH sent to the Protocol Guild please!

@bankless @frogmonkee Ignore the noise. This is when the good stuff gets made 🏗️ #BuildTheDip

Worried about a bear market? Don't be. This is a great opportunity. You just need some survival tips to get through it. @frogmonkee shares them here: newsletter.banklesshq.com/p/how-to-survi…

Ignore the noise. This is when the good stuff gets made. #BuildTheDip

@econoar More dynamic!👌You guys have great chemistry, it's fun to hear candid banter between buds vs more prepared talking points. Welcome back!! 🥳