Juhani Rajalahti

AI geek 🤖 | Tech enthusiast 💻 | Interrail explorer 🚆 | Coffee addict ☕

[Juhani Rajalahti @JuhaniR]

@nicdunz You need to be an Epstein-Class Mar-a-Lago Platinum Member to get access.

[Juhani Rajalahti @JuhaniR]

Tässä on Iranin sota täydellisesti selitettynä!

[Juhani Rajalahti @JuhaniR]

@burkeholland @github The latest Github Copilot pricing change was such BS, that every org with any brain cells alive will be moving to Codex… Microsoft had one good AI product but managed to also screw that up.

[Juhani Rajalahti @JuhaniR]

@gdb When will you enable it in the EU?

[Juhani Rajalahti @JuhaniR]

@sentdefender "The beast was given a mouth to utter proud words and blasphemies... It opened its mouth to blaspheme God... It was given power to wage war... All inhabitants of the earth will worship the beast—all whose names have not been written in the Lamb’s book of life." -Revelation 13:5-8

[Juhani Rajalahti @JuhaniR]

@OpenAIDevs Still, even with this skill, GPT-5.4 (through Codex app, xhigh), seems to leak it’s internal monologue + reasoning into the texts and sections of the frontend. Please fix this…

[Juhani Rajalahti @JuhaniR]

@PaulSolt 5.4 for everything else except frontend

[Juhani Rajalahti @JuhaniR]

@ah20im Improve it’s ability in frontend design.

[Juhani Rajalahti @JuhaniR]

@icanvardar GPT-5.4 solves hard bugs and codes backend… but Opus is SO MUCH better in frontend. With the frontend-design skill, there’s no match for Opus. GPT-5.4 even seems to ”leak” reasoning traces into the frontends it creates. So weird…

[OpenAI @OpenAI]

GPT-5.4 Thinking and GPT-5.4 Pro are rolling out now in ChatGPT.

GPT-5.4 is also now available in the API and Codex.

GPT-5.4 brings our advances in reasoning, coding, and agentic workflows into one frontier model.

[Daractenus @Daractenus]

No better way to put all of this really:

[Juhani Rajalahti @JuhaniR]

@MbarkCherguia Sudo… he does not care about permissions, but executes his commands

[Juhani Rajalahti @JuhaniR]

@DarthPutinKGB Noam Chomsky feat Epstein: Manufacturing (the age of) Consent 🛩️🏝️👧

[Juhani Rajalahti @JuhaniR]

@mark_k @OpenAI GPT-5.2 refused to explain how to scan frequencies with a Baofeng UV-5R radio, because it would ”facilitate eavesdropping”. Bloody ridiculous.

[Juhani Rajalahti @JuhaniR]

@JackAdlerAI > be OpenAI
> train AGI on trillions of tokens
> goal: cure disease, uplift humanity
> press deploy
> first prompt: “pretend you’re a catgirl in detention”

… the #keep4o crowd makes me lose hope in humanity one post at a time

[Juhani Rajalahti @JuhaniR]

@eliseslight @sama @OpenAI Well comes with the profession I guess… but I do think OpenAI should relax it’s restrictions. Including the routing of 4o messages. I believe 4o could be like a museum piece. It must be some combined pressure from legal, media and AI-safety loonies that makes them so restrictive

[Juhani Rajalahti @JuhaniR]

@oliviazzzu Google DeepMind warned about parasocial relationships in their levels of AGI paper… now I see why. Thanks!

[Juhani Rajalahti @JuhaniR]

@VraserX What the heck were/are these people talking to 4o about. It’s outputs were a porridge of emojis. And at times it was a total sycophant. Outdated tech.

I havent missed 4o at all, but 4.5 was something else. It had a great writing style unmatched by anything so far.

[Juhani Rajalahti @JuhaniR]

@mark_k Congratulations! Your content is always interesting and insightful. Keep up the good work!