The Kid Builds

I test AI agents and automations. What worked. What broke. What I kept.

[The Kid Builds @TheKidBuilds]

https://t.co/zF0v8pAVri

[The Kid Builds @TheKidBuilds]

@notmissing_ Sick work brother

[The Kid Builds @TheKidBuilds]

Here is the OpenAI Jalapeño scoop in plain English.

They built their own chip with Broadcom to run models, not train them.

First tests: 1.5x-1.9x more work per watt. 1.7x-3.6x faster answers than the Nvidia systems they compared it to.

Year-end deploy. Small volumes first.

[The Kid Builds @TheKidBuilds]

@markfenner @ViC305 @mr_r0b0t im just a guy wiht a macbook, a Hermes Agent, and too many AI subscriptions

[The Kid Builds @TheKidBuilds]

https://t.co/OoOhurPTNK

[The Kid Builds @TheKidBuilds]

@markfenner im just vibing with my hermes agent and me lmao sneak peak

[The Kid Builds @TheKidBuilds]

The thing worth stealing from an AI voice system is not the prompt.

It is the memory around it.

Approved drafts. Rejected drafts. Modes. Source rules. Corrections. That is what got 58 bookmarks.

[The Kid Builds @TheKidBuilds]

https://t.co/zF0v8pAVri

[The Kid Builds @TheKidBuilds]

@markfenner Thank you. You will love the boat test next lmao

[The Kid Builds @TheKidBuilds]

@notmissing_ I like grok bot because it just works. But my baby will always be Hermes

[The Kid Builds @TheKidBuilds]

@XFreeze I’ve been abusing grok 4.6 such a fun model

[The Kid Builds @TheKidBuilds]

https://t.co/OoOhurPTNK

[The Kid Builds @TheKidBuilds]

@Jaaneek Absolutely been testing it a lot lately https://t.co/Geup5Zy6sw

[The Kid Builds @TheKidBuilds]

https://t.co/OoOhurPTNK

[The Kid Builds @TheKidBuilds]

@cobi_bean Same goal different path 😂

[The Kid Builds @TheKidBuilds]

I thought the "all five AIs picked 17" post was bait.

Then I ran the same test.

ChatGPT, Claude, DeepSeek, Gemini, Grok.

Same prompt. No extra wording.

17, 17, 17, 17, 17.

[Creagen Dow @CreagenDow]

I asked ChatGPT, Claude, DeepSeek, Gemini AND Grok to pick a number between 1 and 30.

ALL FIVE SAID 17.

No coordination. No prompting them toward it.

Five different AI models.

Same. Exact. Number.

What is going on with 17? 😂

[The Kid Builds @TheKidBuilds]

You know @bot is a good product if Mo is actually using it.

[Mo @atmoio]

I didn't expect to actually use this (Grok Bot)

[The Kid Builds @TheKidBuilds]

@atmoio You know @bot is a good product if Mo is actually using it.

[The Kid Builds @TheKidBuilds]

@TheCraigHewitt Same but I’m finding grok writes way more code.

[The Kid Builds @TheKidBuilds]

https://t.co/OoOhurPTNK

[The Kid Builds @TheKidBuilds]

@yeahfortommy I just did a test myself and just switched to supergrok plus instead of OpenAI pro

[The Kid Builds @TheKidBuilds]

https://t.co/OoOhurPTNK

[The Kid Builds @TheKidBuilds]

@TENTHDEMENSION Man that sucks I’m doing a test right now with @Da7_Tech boat prompt

[The Kid Builds @TheKidBuilds]

@JosiTonedice Lmao same I was looking at inter arc cards

[The Kid Builds @TheKidBuilds]

M6 mini from $899. M5 Ultra Studio from $5,499.

- Mini: M6, Dual 16-core Neural Engine, up to 64GB
- Ultra: up to 36-core CPU, 80-core GPU, 1.2TB/s memory
- Both ship September 22

This is how Apple wins. They sell the local AI box. They do not need a model.

[The Kid Builds @TheKidBuilds]

@darekgusto I’ve run Gemma 4 and I agree they can’t compete for now

[The Kid Builds @TheKidBuilds]

@darekgusto I think the new fade is local inference so that’s why these will sell

[The Kid Builds @TheKidBuilds]

@JosiTonedice 😂😂😂