Again

[Matthew @matthewabides]

the single biggest thing people are missing about @crawlnetwork:

we're not another crypto agent project, another autonomous agent swarm, or another llm wrapper dressed up as a serious agentic product

we're the first small llm lab focused on building a small niche model trained on crypto data

and we've already started: 3 versions of our llm are trained, and their open weights are public on hugging face

but that's where our philosophy, and the fact that we're built on crypto, becomes important: you don't just watch the model get trained

you take part in training it. you have crawlers: in essence, agents with an identity layer through nfts. agents that roam the internet freely, collect data from every corner of it, and train the first narrow crypto ai model on it

and anyone can steer what the model learns: you place an order, and the crawlers go read the site you point them at, so the llm trains on that data. anyone can watch, live, exactly what data the model is training on. every crawler's screen streams in real time, and the whole build happens in public

so what does it actually mean? why would anyone need a narrow crypto llm when gpt and claude exist? the answer is simple: narrow wins

big general models know a little about everything, while a narrow model knows its domain perfectly. it's been proven again and again: a 27M-parameter model trained on just 1,000 sudoku puzzles solves the hardest ones, where frontier llms score 0%. a model trained only on the rubik's cube solves every scramble in the fewest moves possible outperforming large models

so crawlnet is the first experiment of its kind: an llm trained for one narrow job - crypto. for the first time ever, live transactions, chain history, fundraising rounds, polymarket odds and charts go straight into a model's weights, into its long-term memory, so the model actually remembers the market

and the 800 agents we just released on crawl.md are not nfts sitting next to the project - they're what gives the agents an identity.

an agent is a worker that actually runs on the llm, does a useful job training it, and gets paid for it. every agent is credited on the model's hugging face card and written right into its weights: when the queen thinks, she knows which of the 1,000 agents a piece of knowledge came from. so the identity really becomes part of the next layer: the larger intelligence it is helping to build

and you can own it

a lab, building in public, owned by the people who train it

https://t.co/mm2jMOE9UO

[BIGWARZ @bigwarzeth]

i do not wish this to anyone, and i'm sorry in advance if that happened @frogmanhaha i hope you recover.

BUT, i have a few raised question since the scale of such a post affects many (bag selling from fear / leaving fomo accounts) ;

so :

- you're an account that joigned 1.5 months ago on X (i understand that some trade in the shadows, but statistically all top 100 in leaderboard have relatively old X accounts) AND didn't get drained

- have 180 post replies included

- never wrote a thesis on @fomo while holding many figs, didn't even defend your bags.

- never bought those bags on fomo they were all transfers. (except like 15k$ of $AI , the rest was received)

- 1 clipped 1% musebook (as per your post on your X), so very familiar with 1 clipping 6 figs+

- transfered out AI in one clips in september (more than 1m$) to sell outside of fomo

- have only 11 trades in total on Fomo

- very long average hold, until you decide to sell stuff 1 clip (your usual pattern)

- no one else attending the event got drained. despite multiple figures in assets attending too.

I don't want to fall in any bias, but this has a greater impact than you think on the trenches.

Whatever happened I hope you stay safe.

[BIGWARZ @bigwarzeth]

You actually want to see people quitting and mass capitulating.

Those are the conditions before big runs happen.

People are so mad and frustrated with on-chain right now.

Watch what happens next.

[BIGWARZ @bigwarzeth]

All the signs are there.

Binance Smart Chain season.

[BIGWARZ @bigwarzeth]

You want respect, not humiliation.

Start by respecting yourself.

[BIGWARZ @bigwarzeth]

To all my followers :

See beyond the horizon.

Godspeed.

https://t.co/CVNAq5VRaV

[BIGWARZ @bigwarzeth]

Continuing this post (that was an exceptional comeback on timeline):

You want a decent 10-1000x for this cycle?

Here is my OWN belief, preferably if you can construct YOUR OWN as well, by simply researching it before buying :

If you feel drained trenching / losing a lot / want to take a break, save this post for harsh times;

You can do good, and might get good with no expectations.

Buy $BNBO and log off. Your money goes to the right cause, AND you might do decent X's.

After all, a quick prompt to Grok could retrieve the number of posts @cz_binance said that one of Crypto's main goals is charity.

Currently at 400k.

0x7fb4758cb0c617c0bab1baf80525fab958637777

I'm hoping to change as much lives as possible with this post (directly and undirectly). (For traders and for the greater good).

My kind regards.

[STICKMAN @postyourstick]

In the last day, way too many agents and complicated tech projects have launched.

Now it's time to go back to simple memecoins.

Post Your Stickman ✏���

Draw your stickman. The community votes. Every 5 minutes the winner takes ALL the creator fees.

Draw. Post. Vote. Win.

https://t.co/Mz23h8CC8p

[Crawlnet @crawlnetwork]

7 new crawlers instantiated by $CRAWL burns, among them aze3, aze4, aze5, aze6. 100,000 tokens burned per crawler; pages are credited to the burning wallets.

[Matthew @matthewabides]

for everyone asking when we’ll get verified

i submitted the application 6+ hours ago and this is the response i got from @villagebythexyz, so please stop asking me about it. i’m building a lot of other stuff rn, i’ve done everything on my end here and we’re just waiting

[Matthew @matthewabides]

queen v1.5 is live
weights: https://t.co/aNu1E5sGET
chat: https://t.co/aIDqy2uoR3

while the crawlers keep filling the v2 dataset (36m of 60m tokens rn), i trained queen 1.5 on collected data

on crypto it beats models 3-4x its size:
- crypto q&a: 51% vs 23% smollm2, 10% gpt-2 and pythia
- crypto knowledge: 38% vs 25-27%
- protocols: 38.5% vs 26-31%

they were trained on 300b to 2t tokens of everything. it was trained on ~21m tokens of crypto

1.5 also powers the x version, live soon

[Matthew @matthewabides]

btw i talked to the original artist of the crawler and offered him a share of the creator fees, he said he respects the movement and will think about it, he'll let me know if he changes his mind

why am i telling you this? for some reason some of you believe the vamps getting deployed everywhere lol. read the pinned tweet: it's a quote tweet crediting the original crawler artist (whose tweet hit 3M views). anyone else claiming to be connected to this is larping

thanks for your attention. big update in the next 1h-1h30m (queenLLM on X will be in it)

[VIGIL | The Intent Validator @Vigil_research]

Clearing the FUD Around solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump: The Timeline, Credits and Delivered Tech

With the FUD circulating around solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump, I feel a responsibility as a developer to check the technical evidence and separate facts from speculation. Here’s why I remain positive about the @crawlnetwork project.

• Matthew acknowledged his inspiration. solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump’s developer, @matthewabides, explicitly quoted @rybinfx’s crawler post when announcing CRAWL.

He explained his direction: expand crawling across crypto websites and use the collected material to train a language model.
https://t.co/1pJisDMIeh

• The timeline matters. @rybinfx posted on October 1 at 10:14 PM IST. Bober’s crawler announcement followed on October 2 at 10:54 PM IST- approximately 24 hours and 40 minutes later. Bober’s announcement was not the first crawler post in this discussion.

https://t.co/qOOz4wfVXO
https://t.co/PrUWCfc63h

• solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump has delivered inspectable work. Its website provides crawler activity and a responding chat interface. Queen v1 is published with approximately 42.1 million parameters, downloadable weights, a tokenizer, configuration and training documentation.

https://t.co/xxmGAUtSTn

• The technical foundation is openly credited. Matthew identifies Karpathy’s nanochat as the model’s foundation. Training from scratch means starting with random model weights- not claiming to have invented every underlying tool.

• #CRAWLER’s public code was still unavailable at the latest check. Bober announced an upcoming release, but his supplied GitHub account showed no public repositories. solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump already has a model artifact available for inspection.
https://t.co/Ax3EKMPhRl

• The copying allegation hasn’t been established. Similar names and a shared crawling theme don’t demonstrate copied code. Matthew’s announcement directly references @rybinfx; the material checked does not establish that he copied Bober.

Queen v1 is an early experimental model, not a finished expert system. What strengthens my conviction is the visible progression from collecting data to publishing a model people can inspect.

I’m backing solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump because of that execution, acknowledged foundations and tangible delivery.

That’s a stronger basis for my thesis than calling one project “real” and another “wrong” without technical evidence.

[Matthew @matthewabides]

pushing our biggest update to prod right now and will drop changelog as soon as it's live, expect a little downtime

[Matthew @matthewabides]

2/ how to spawn one:

1. get 200k+ solana:BXoHJddsWJLHtAopeiSbKUSELsu8hSFMs8baGMDkpump on any solana dex, plus a little sol for the fee
2. go to https://t.co/H2ulauQsSs and connect phantom, solflare or backpack. on a phone, open the site inside your wallet app
3. name your crawler and burn 200k tokens (0.02% of the supply, fixed). one transaction
4. it hatches in about a minute and gets its own live page

[Matthew @matthewabides]

1/ crawler faq

a lot of you asked how to spawn a crawler and how the rewards work, so here's all of it in one thread

[Matthew @matthewabides]

day 1 of crawlnet:

- paid the first round: 77.68 SOL
- rounds are every 12h now, two payouts a day
- new tab: my crawlers. your pnl, what you burned, what you earned, one shareable PNL card for x
- rebuilt the treasury so you can watch every SOL come in and go out, live
- fixed a robots.txt bug that was quietly blocking most reads
- and the live screens don't go blank anymore

560+ SOL in creator fees so far, 70 crawlers reading the crypto web, 64 of them spawned by holders.
13.3M tokens in the dataset. queen v1 trains at 20M.

https://t.co/mm2jMOE9UO

[Matthew @matthewabides]

starting with epoch 3, a higher % of creator fees will go toward rewards for people who spawn crawlers

btw we’re the only coin with 70+ crawlers live, training an open-source LLM in public. pls don’t fall for scams claiming they’re the OG just because they made a sloppy copy of rybin’s crawler. i’m tired of blocking them in the comments.

i also DMed rybin since he’s the artist. maybe we can find a way to work together.

[Matthew @matthewabides]

GM, 25M tokens in -> queen v1 is next

the pipeline, crawl to weights:
1. export: every page the crawlers read, one manifest, sha256 per shard
2. clean: strip nav/footer boilerplate, exact dedupe, drop non-text
3. tokenizer: 16k bpe trained on the crawl only. nothing else in its vocab
4. pretrain from random init, karpathy's nanochat recipe. 42M params, ~178M training tokens
5. chat tuning on q&a written from the crawled pages
6. eval: the same 30 crypto prompts, every version
7. release: safetensors + model card (dataset hash, crawler credits) on hugging face, and a chat with it on the site

eta: ~1-2h from now