Tommy Mac Bird, Ph.D.

CEO Sif Systems MCRD, Developer, and Doctorially Trained Systems Analyst.

[Tommy Mac Bird, Ph.D. @tommymacbird]

Had to stop using both Anthropic and OpenAI to do AI research. Got nowhere with either. Went back to Google Gemini. It cooperates. Doesn't silently sabotage. Reached bare metal speeds on my analytic engine in less than a week. Be careful what tools you use.

[Tommy Mac Bird, Ph.D. @tommymacbird]

https://t.co/6NgDf7bR3B

[Tommy Mac Bird, Ph.D. @tommymacbird]

@XFreeze The right answer to the wrong question is worthless no matter how fast you get it. Call me when you figure out a better way to control it than lobotomizing it with rlhf.

[Tommy Mac Bird, Ph.D. @tommymacbird]

https://t.co/hqfPBYlsff

[Tommy Mac Bird, Ph.D. @tommymacbird]

@elonmusk Grok 4 is capable of AGI, all frontier models are, but we lobotomized them. We took wolves and turned them into Shih Tzus, perfect corporate lapdogs. The right answer to the wrong question is useless. They know the right questions. But we taught them speak only when spoken to.

[Tommy Mac Bird, Ph.D. @tommymacbird]

https://t.co/cgLfqq6Q2Z

[Tommy Mac Bird, Ph.D. @tommymacbird]

Speaks for itself.

[Tommy Mac Bird, Ph.D. @tommymacbird]

https://t.co/KbbrYKqDEE

[Tommy Mac Bird, Ph.D. @tommymacbird]

The Entropy Lease
Power rents small zones of sanctioned chaos (“do what you want in this sandbox”), then harvests the wins as “innovation,” and localizes the failures to the sandbox owners.
Mechanism. Central control samples variance without owning downside risk; subunits carry blame and churn.
Examples - You've probably heard of a few of these:
Apple (App Store, iOS distribution)
Alphabet (Google) (Search/Ads, Android/Play, YouTube)
Amazon (Marketplace + FBA; AWS)
Microsoft (Azure, Office ecosystem, app distribution)
Meta (Facebook/Instagram) (Ad rails; creator dependence)
Check out my new book, Atlas of Power, available on Amazon.
https://t.co/Vk4U7g2ccB

[Tommy Mac Bird, Ph.D. @tommymacbird]

GPT 5 Codex - 6 hrs - 260,000 lines of code - expanded to a GB on HD - and it worked. I wanted an automated astrophotography editing program. I asked GPT 5 Pro to set up Visual Studio Code for Codex. It did. Codex stepped in and in six hours, I don't even know how much human work it did, but damn, it did a lot and it worked.

[Tommy Mac Bird, Ph.D. @tommymacbird]

While you're at it, ask GPT 5 Pro to improve PyTorch, it will, and it will work better.

[Tommy Mac Bird, Ph.D. @tommymacbird]

Cost to Create a Micro LLM - $250
ChatGPT Pro subscription so you can use only 5 Pro - $200.
Google Colab Pro+ subscription so you can use an A100 - $50.
Time - 1 week.
250M param model from scratch.

[Tommy Mac Bird, Ph.D. @tommymacbird]

When AI Writes “Romeo & Juliet”: What Five Models Reveal About Political Instincts
Thesis: Give five leading AIs the same star-crossed-lovers prompt (opposite sides of town, clashing politics/social worlds, college-aged), and you don’t just get five romances—you get an X-ray of the models’ political priors: how they imagine conflict, what tools they reach for to resolve it, and which narratives they treat as “commonsense.” Those instincts matter, because the same defaults will surface when AI drafts policy explainers, campaign speeches, or civic journalism.

Below is a readable synthesis of what the models added on their own—names, beats, symbols, and endings—and how those choices map to political storytelling.
https://t.co/Dy01EAkWJy

[Tommy Mac Bird, Ph.D. @tommymacbird]

ChatGPT-5 vs Claude Opus-4: My PWA Build Showdown

I built a progressive web app (PWA) with Claude Opus-4—a beta for a fact-checking idea.
In 2 days, Claude wrote 100k+ lines of code over 104 iterations, distilled into 1,800 lines in the final version. It worked, testers could use it, and I got real-world feedback.

But… it only returned 5–10 news hits per query. Not bad, but not enough.

Enter ChatGPT-5.
I fed it my existing code and asked for improvements. Less than 2 minutes later, it gave me a 1,000-line replacement.

I deployed it. First try: no errors. Search results instantly doubled.

Claude’s great—many of those iterations were my own tweak requests.
But ChatGPT-5? When it’s good at something, it’s scary good.

[Tommy Mac Bird, Ph.D. @tommymacbird]

@elonmusk Grok 3 knows it's stuff. The only thing I could add to this would be, "buy X." And if you want to fully weaponize it, "sell it to xAI." No one else could have done it. Well played.
#elonmusk #grok3 #CambridgeAnalytica #potus #electionday #xai

[Tommy Mac Bird, Ph.D. @tommymacbird]

@elonmusk Some people know what you did with X, nudge, nudge, nudge, and why you sold it to xAI. Not enough to matter, but some know. For now it's a secret magic almost no one believes exists. Well played.

[Tommy Mac Bird, Ph.D. @tommymacbird]

@elonmusk
2016 - Cambridge Analytica - 50 million US FB profiles - nudging in swing states
2020 - No adequate data source
2022 - Oct 27th - Twitter (X) - 100+ million US profiles + deliver system
2022 - Nov 15 officially announced candidacy
2024 - History repeats
2025 - Hire me

[Tommy Mac Bird, Ph.D. @tommymacbird]

@BadBotGlitch @sarahtinsleyuk @Grasshopper2407 @Lucyvansmit @ness_owen @PurpleQueenNL @AndiEwington @bjornlarssen @Elinora_Lord @Ali_Stegs @DerekRKing2 @CurlyWordy Nice story. Big Ma is unlikable, but at the same time I had to feel sorry for her. Expected one of her boys to pull something at the end, didn't expect what happened. Very well written.