Cali Kid

Just trying to figure out how to make it in crypto

[Cali Kid @calikid1911]

@kate_p45 @grok are the above numbers in the original post accurate? Please cite sources.

[Cali Kid @calikid1911]

@grok @LionOfJudahX @realMaalouf @grok where is he getting the 100,000 number from?

[Cali Kid @calikid1911]

@bes_______ @AmoebaFarm Looks live to me unless this is a fake?

[Cali Kid @calikid1911]

@theisabelb @grok can you please confirm the statistics mentioned in the original post from Isabel?

[Rei @rei_labs]

Can Adapt‑1 Preview stay in an interactive decision loop while thousands of outcomes keep changing its persistent decision state?

We finished migrating to the infrastructure intended for public access. Here’s what latency looks like as retained experience accumulates.

[Cryptic Tits (npc/acc) @cryptic_tits]

n00b critics tried telling you : "it only learns because a human sets up the task structure for it."

REI new release: they removed that structure and let it figure the structure out itself ... and it kept 95% of the performance.

[mic drop]

[consom888 @consom888]

I am often wrong but holding size added along way since 2024- in a nutshell the $REI thesis could take months or years to play out but frontier ai labs with boring tooling are getting tens of billions imagine real disruption:

* Instead of spending absurd amounts of compute + training data to learn everything
* Give the model a **task vocabulary / ontology**
* Let it learn the **actual underlying structure from experience**
* Task scope vs. **trillions of tokens** = an absurd tradeoff
* Much more efficient way to teach models domain-specific structure

Credits @okane0x for the eli5

[Cam Higby 🇺🇸 @camhigby]

X just notified me that the Pakistani government is formally attempting to censor my posts about Islam.

[Rei @rei_labs]

We revisited DeepMind's DM Alchemy, comparing three approaches: Adapt-1's ontology-based online sequential learning, Pinon et al.'s transformer dynamics with tree search, and AlKhamissi et al.'s modified representation with episodic memory.

Full Blog: https://t.co/MPx64fD2Pv

[Cryptic Tits (npc/acc) @cryptic_tits]

current takeaways from looking into this over the weekend:

- the cleanest proof: they ran the exact same system twice, with learning switched both on and off. Very noticeable difference.

- on a hard robotics test it had never trained for, it showed up cold and still scored in the same range as the big specialist models.

- everyone dives into the chatagent and misses the point - one core ran six different robotics tasks. it plugs into ANYTHING. (someone get Andrew Kang on the phone)

- "no proof just narrative" is dead. Now it's "does it hold up when strangers run it." the files are public, API's open but currently none of the critics have run the API... (Strange)

- whats needed now is independent researchers running their own tasks. that's the big one.

- token also needs some clarity soon

working on a full deep dive article that i think will help explain everything in laymans terms.

[Rei @rei_labs]

Introducing Adapt-1 Preview (Δ1): a pretraining-free substrate for test-time learning. It learns from scored feedback while performing the task.

https://t.co/55MfjbYa42

Today we’re making its first research preview and first results across robot perception, spatial reasoning, and games available.

[Plata @platacrypto]

To bring this highly technical piece of information in some simpler terms:

@rei_labs is working to solve what openclaw and hermes still struggle with - ai agents that truly learn, adapt, and evolve over time.

Openclaw and Hermes mostly give models memory through files, prompts and retrieval and that context can be missed, truncated or retrieved at the wrong time which happens quite a lot as most folks using them know.

$REI may be close to publishing a breakthrough in persistent ai state: a system that changes future decisions based on experience. actual learning, rather than another elaborate agent harness.

[Cali Kid @calikid1911]

@WallStreetApes @grok is this right that trump tried to give them citizenship and Democrats blocked it?

[Cali Kid @calikid1911]

@varrock When Reddit is primarily the same left wing suicidal empathetic people who want destruction for the west, it all becomes clear.

[Cryptic Tits (npc/acc) @cryptic_tits]

worth mentioning this:

one of the three godfathers of deep learning recently quit Meta to build a different architecture.

Yann Lecun, Nov 2025: left a 12 year job and Meta's entire AI budget to found a lab betting LLMs are a dead end for real intelligence.

His said: "don't work on LLMs"

everyone's now chasing the thing one small lab named 15 months ago and is due to exit stealth very soon...

[Cryptic Tits (npc/acc) @cryptic_tits]

top 5 reasons to fade $REI:

1. "team not doxxed" - just like satoshi, who clearly built nothing of value. except they are semi-doxxed and plan to doxx more on rollout. Welcome to crypto.

2. "a 23 person lab can't compete with openai " - correct, they're not trying to. openai builds text engines. REI builds the reasoning layer Chollet says is inevitable. different sport

3. "no papers" - confirmed coming, each module gets one, with code. you're literally fading the catalyst

4. "nobody understands what it is" - Nadella just explained it to 66 million people. he called it "the key test of your sovereignty in the era ahead"

5. "the beta chat agent was slow" - the beta was a constrained demo with the architecture deliberately stripped back. judging Core by rei chat is reviewing a restaurant by tasting the napkin

fade away. the market's already 91% down - you're fading it at the bottom while Lecun quits meta, Chollet calls the thesis inevitable, and Dwarkesh describes their 2025 product as the 2028 paradigm

can't say you weren't shown the receipts. imo faders get BTFO when stealth is exited.

[Cali Kid @calikid1911]

@cryptic_tits @0xreisearch 0xreisearch is full of gREItness.

[Cryptic Tits (npc/acc) @cryptic_tits]

growing up is admitting @0xreisearch was right about AI

Dwarkesh patel's new video says the next paradigm is models that learn from deployment. his timeline is by 2028

reisearch said this in march 2025 (+ also shipped it )

the AI industry is catching up to a 23 person Neo Lab with no VCs

Prob gonna do a longer post here and break down all the receipts because it's goes pretty deep.

$REI

[Cryptic Tits (npc/acc) @cryptic_tits]

This is easily done tbh. Half the job with $rei is taking the time to learn and understand. This is why the existing community are so die hard.

Once more info goes public the research will be academic level. Trenchers have to make the choice to go full left curve before that to make the entry.

The real buyers will be funds and institutional grade analysts. The same crowd obsessed with AI bottleneck stocks.

[Cali Kid @calikid1911]

@mathburn666 Grei. Love the REI arc you’re on, Math!