jefflau.eth

Co-founder @ensdomains. Distilling the complicated into the simple. In local AI psychosis.

[jefflau.eth @Jefflau]

Thread of my threads distilling the complicated into the simple.

Previous topics include:
- Zero Knowledge Proofs
- EIP-1559
- Rollups
- ETH2 and the merge

I’ll add new threads here for easy referencing for myself and others.

[jefflau.eth @Jefflau]

@yazins Longstartupdomain is still a token that an ai has to find. How do you expect the model to find it and weight it heavier than other tokens?

.com is actually forever in the model weights historically so i can see ai weighting it heavier when it does its own web searches.

[jefflau.eth @Jefflau]

Hard disagree!

AI means that machines can finally read and understand, many domains, sites, apis will be made by AIs themselves and will be important. And domains will be even more important to cut through the noise. How does an ai find your service? Through a domain name, not an ip address that may change when you migrate from railways to aws. Of course I’m very biased - but is my real opinion!

[jefflau.eth @Jefflau]

All the justification I need

[けーず @llm_frontier_jp]

DGX Spark 100万円 ←まあ高い
RTX 5090 120万円 ←高い
RTX PRO 6000 Blackwell 300万円 ←高すぎ
DGX Station GB300 2000万円 ←高すぎいいい

人が足りないな。雇おうか。

1人雇う → 500万円/年
3人雇う → 1500万円/年

もう一度GPUの価格を見る。

DGX Spark 100万円 ←安すぎ、10台買おう
RTX 5090 120万円 ←安すぎ、10枚買おう
RTX PRO 6000 Blackwell 300万円 ←安すぎ、3枚買おう
DGX Station GB300 2000万円 ←うん、買うかな

[jefflau.eth @Jefflau]

Your point is that the model makes it more reliable? My point is that is the product harness makes it more reliable

I agree with the second statement. It’s not a competiton between grokbot and dots. Its that the bot level product is an iteration of the harness and makes the experience better (no matter which current product is ahead)

[jefflau.eth @Jefflau]

Product beats models at this point

[Mia @MiaAI_lab]

Grok Bot is unstoppable at this point.

[jefflau.eth @Jefflau]

@gregskril Also this isn’t targetted at dots vs grokbot, its that the product you build around the model is more important than the model itself.

[jefflau.eth @Jefflau]

@gregskril Skill issue

[validator.eth @ValidatorEth]

Creating DNS look alike usernames is a user safety failure, and a completely avoidable one. We keep repeating a mistake this industry has already paid for.

[Erik Voorhees @ErikVoorhees]

Crypto grew up in the Dark Forest, and the entire world is about to find itself there

[jefflau.eth @Jefflau]

My first ever X subscription to some mad smart brit doing amazing things with localai!

[Ash Hart @ashxhart]

We've got our first subscriber!

Thank you @FurterAlain for backing me and TensorFold. Really appreciate you being the first to jump in.

I'm putting together two substantial subscriber only pieces each month, with practical local AI setups and a closer look at what goes into building TF. I want you to come away with things you can actually use on your own setup.

I will also be setting up a private group chat for subscribers.

[jefflau.eth @Jefflau]

@ashxhart @FurterAlain Subbed!

[ens.eth @ensdomains]

Are you ready?

[jefflau.eth @Jefflau]

@MiaAI_lab @ashxhart How does official weights on tensorfold work?

[jefflau.eth @Jefflau]

@MiaAI_lab Can’t wait for this one!

[jefflau.eth @Jefflau]

Mia cooking 👩‍🍳

[Mia @MiaAI_lab]

DeepSeek-V4.1-Flash EXL3-EAQ v1.1 is out 🧪

What's different vs v1?

Sparse attention uses an indexer to pick which past tokens matter. v1.1 keeps that whole path in full BF16 instead of quantizing it (the indexer, shared query projection, and compressor).

When fed the original model's inputs, v1.1 makes the exact same picks 100% of the time past 16k (403,560 / 403,560), vs 94.95% on v1.

In a full quantized run, both land near 73%. That's because the inputs reaching the indexer have already shifted from quantizing the rest of the model, so a BF16 indexer alone can't fix that gap.

v1.1 weights just ~0.15 GB more than v1.

Please note that this is experimental ‼️

Get it here:
https://t.co/yMmNvpb4gj

[jefflau.eth @Jefflau]

@EnsBtc @nvidia @ashxhart @MiaAI_lab Run local ai. Keep your data private. Sovereign intelligence

[jefflau.eth @Jefflau]

I've already convinced more than 1 person to buy a cluster of @nvidia DGX sparks and following @ashxhart and @MiaAI_lab to start their local AI journey.

And so the evangelism begins...

[jefflau.eth @Jefflau]

@sudoingX how?!

[jefflau.eth @Jefflau]

@MiaAI_lab Looking forwars to it