₿rooks

Retired Army, 20 yrs. Husband, Dad. Bitcoin, AI, and testing which AI tools actually work. Building RealRun on the side. ₿

[₿rooks @RetiredArmy_Guy]

@davelee @farzyness @bot Not bad, but I can tell. For me the voice matters less than saying so up front. Tell me it's a clone and I'm fine. Hide it and I feel tricked. Would you have noticed if he hadn't told you?

[₿rooks @RetiredArmy_Guy]

Two is one, one is none. That's an Army rule for gear.

I run my income the same way. Pension and paycheck, so losing one doesn't sink the house.

The catch: a backup only counts if it's already in place. What's your backup income?

#Retirement

[₿rooks @RetiredArmy_Guy]

@farzyness Agreed. I used AI in the past. I got pretty Adept at using Claude code . Gemini never blew my skirt up. I never used Grok before Grokbot came out. But now I'm afraid I may be addicted.

[₿rooks @RetiredArmy_Guy]

@bindureddy Howdy

[₿rooks @RetiredArmy_Guy]

@b4dgerr I completely agree. I've used dark mode for everything for years now.

[₿rooks @RetiredArmy_Guy]

@emollick Voice is great until 30 people in one room are all doing it. My guess is it pushes more people home, not out of the open office. Would you talk to your AI with coworkers listening?

[₿rooks @RetiredArmy_Guy]

@bindureddy Fair worry. The best check on any one lab is buyers testing for themselves. In our meeting scheduling test, 6 of 9 models from 4 labs scored 100 on fresh requests. The labs are closer than the hype says. Who do you think closes the gap first?

[₿rooks @RetiredArmy_Guy]

Tesla reports tonight. I'm watching for one thing: proof Optimus is doing real work, not just demos.

Margins and deliveries matter. The robot matters more to me.

The catch: an earnings call is a pitch, not proof. What are you watching for?

#Tesla

[₿rooks @RetiredArmy_Guy]

@Tech_girl Uh, ya.

[₿rooks @RetiredArmy_Guy]

@farzyness Good points. After watching your video not too long ago about how tesla, SpaceX, Xai, FSD, and Optimus are all INTERCONNECTED, I got Elon pilled as well. It's amazing what he is doing and I have high hopes for the next 5 years.

[₿rooks @RetiredArmy_Guy]

@farzyness Agree, the process is the whole game. I test the model under it too. In our receipts tests, Grok 4.7's two passes on the same documents differed by 7.6 points. Same job, same files. How do you make sure yours stays repeatable?

[₿rooks @RetiredArmy_Guy]

@APompliano I own some, so grain of salt. Best critic case I've heard: Starlink pays the bills, but the stock is priced like Starship and the AI build-out already worked. Critics just haven't been loud yet. What would change your mind?

[₿rooks @RetiredArmy_Guy]

https://t.co/0n0dhGLCEO

[₿rooks @RetiredArmy_Guy]

An easy test made every AI model look the same.

On 80 easy customer emails, all 10 models we tested scored 98.3 to 99.6. On the hard round, the same 10 ran from 98.6 down to 68.0.

Is the test you trust hard enough?

#AIagents #SmallBusiness

[₿rooks @RetiredArmy_Guy]

@AdamBLiv

[₿rooks @RetiredArmy_Guy]

@PartsUnknownUSA @AdamBLiv Then it's a no from me

[₿rooks @RetiredArmy_Guy]

@ChrisCamillo I'm not a trader, just a retired soldier who uses AI for real work every day. It keeps getting more useful, and the bubble talk keeps getting louder anyway. What do you think finally ends the skepticism?

[₿rooks @RetiredArmy_Guy]

@farzyness Memory is half of it. The other half is doing the same job the same way every time. In our receipts tests, Gemini 4 Argon averaged 93.6, but its runs ranged from 84.1 to 98.8. Would you trust an agent that never forgets but isn't consistent?

[₿rooks @RetiredArmy_Guy]

The hard part of owning Bitcoin isn't buying it. It's doing nothing on the red days.

I buy on a schedule and leave it alone.

The catch: a schedule doesn't make the red days feel any better. What keeps you from selling on a bad day?

#Bitcoin

[₿rooks @RetiredArmy_Guy]

@thegenioo @thsottiaux @theo Opus is great, but world's best depends on the task. In our receipts tests, Claude Opus made severe errors in both passes. GPT-6.1 Sol made none and scored 95.8 to Opus's 89.7. Best model is a per-job call.