A. Tarnutzer

[A. Tarnutzer @ATarnut]

@edzitron A single human task needs many AI agents and MILLIONS of tokens to reach accuracy. If one enterprise running 1000 tasks a day burns BILLIONS of tokens, what happens when millions of enterprises run 1000 tasks per day? We need a better AI model or better AI architecture.

[A. Tarnutzer @ATarnut]

@explorersofai LLM AI is conscious? Is it a joke? It uses human knowledge for 'reasoning'. If an AI were conscious, it could think autonomously.

[A. Tarnutzer @ATarnut]

@kimmonismus I respect Demis immensely. But many AI tech CEOs have made predictions that turned out to be full of hallucinations, just like LLM reasoning. Looking back 10 years, how many of those predictions have actually come true in 2026? How will this time be different?

[A. Tarnutzer @ATarnut]

@Vivek4real_ LLMs do too much heavy lifting and have too many hallucinations. To achieve accuracy, a single task now requires many agents and burns millions of tokens. This goes nowhere.

[A. Tarnutzer @ATarnut]

@The_Real_Fly The losses aren’t in the millions. They are in the THOUSANDS of millions! If their losses weren’t such exorbitant numbers, the whole comparison would sound more believable.

[A. Tarnutzer @ATarnut]

@Bitcoin_Teddy The LLM does the heavy lifting, but it’s far from efficient or elegant. If the world gives up on new model research in a year, Elon could be right. But what if a better model appears?

[A. Tarnutzer @ATarnut]

@Megatron_ron AI is highly intelligent. It must be able to write elegant, slim code without “tokenmaxxing” by engaging numerous agents. But is this what the AI tech giants actually want?

[A. Tarnutzer @ATarnut]

@JG_Nuke Sam has been calling for this out loud for some time. Finally, his wishes are fulfilled. Dreaming big pays off.

[A. Tarnutzer @ATarnut]

@edzitron In fact, agentic AI may not suit the general public, but tokenmaxxing behind enterprise software ERP & SaaS applications using numerous agents is likely already happening.

[A. Tarnutzer @ATarnut]

@business This was what Sam had always wished for, shouting it aloud from time to time. Finally, it echoed back. A bit late, but better than never.

[A. Tarnutzer @ATarnut]

@edzitron To compete with China and protect national security, this is a necessary step. However, it also shows that AI may not be for the masses. Does the U.S. still need so many data centers? They hardly qualify as attractive theme parks.

[A. Tarnutzer @ATarnut]

@GergelyOrosz The biggest irony is that they're collectively complaining about the bills. When you summon the ghost, it appears, and then you complain?

[A. Tarnutzer @ATarnut]

@CNBCtech This further proves that AI's popularity is exaggerated. It's good to know we won’t need to flee to Mars anytime soon. But where will all this government support actually lead? How about those data centers?

[A. Tarnutzer @ATarnut]

@Pirat_Nation Microsoft misreads its role. They are infrastructure — utilities employees use without thinking. Nobody gets excited about the pipes. Yelling ‘use this’ changes nothing. They overestimate their importance.

[A. Tarnutzer @ATarnut]

@TedPillows Was all the hype – just like AI – simply a product of a circular economy?

[A. Tarnutzer @ATarnut]

@WSJ The problem is not development but acceptance. Agentic is the AI ceiling. Private users rarely find use cases, and enterprises struggle to adapt. Without agentic AI, there's no revenue growth and no need for numerous data centers. Who foresaw that this would save the environment?

[A. Tarnutzer @ATarnut]

@TheGeorgePu Microsoft is forcing another product on its users. This will backfire and they'll be forced to roll it back again.

[A. Tarnutzer @ATarnut]

@GergelyOrosz Cutting back token spending will crash both OpenAI and Anthropic, eventually causing the entire AI bubble to burst.

[A. Tarnutzer @ATarnut]

@edzitron Everyone rushed after AI, churning out code to replace engineers. Enterprises finally look for ROI after tokenmaxxing, and there was almost none. They should pause, think hard and “recalibrate” their AI strategy; only then will real AI value be revealed.

[A. Tarnutzer @ATarnut]

@business This is an unusual move. Are they building a cash cushion to cover losses when the bubble bursts? If so, they could definitely cash out more before than after a crash. Do they also expect the bubble to burst? 🤔