It's over.
In a Nutshell
Anthropic's new Fable model is crushing competitors and dominating the enterprise API market, prompting the US government to intervene with demands to slow down or pull the model under the guise of safety concerns. The speaker dismisses the safety claims as political theater driven by rival labs and power dynamics, comparing the situation to a messy Game of Thrones power struggle rather than legitimate regulatory oversight. He concludes that current AI models remain fundamentally limited tools requiring human direction and are nowhere near replacing engineers or creating production-ready code without oversight.
These notes were generated by AI and may contain inaccuracies.
The speaker played extensively with Claude Fable on Friday and found it to be a really good model. They did not examine any code during testing, focusing instead on quickly creating games. The speaker has moved away from productivity apps entirely, stating that productivity tools are dead because anyone can now build their own. They noted that their entire career of making productivity apps could now be rebuilt in 15 minutes by others.
The speaker recommends moving on to making games and made three to four game concepts on Friday morning through afternoon. They found nothing they could throw at the model that it couldn't handle. Anthropic's $200 plan is described as extremely generous, providing $8,000 worth of usage according to SemiAnalysis.
The current state of AI has become a political game for the giants and titans of industry. When playing at that level, the rest of us only receive a trickle of news from sources confiding in journalists, which likely involves exaggeration depending on the source. Information from Anthropic will differ from information from the US government.
The speaker dislikes politics because it is an information asymmetry game where peasants lack insider information needed to form accurate assessments. When in doubt, assume actions are about power and capital. Safety has never mattered to anyone. The idea that the US government got spooked safety-wise by Fable is dismissed as not having happened.
Anthropic has gained too much power and is kicking everyone's ass, running circles around other labs. Other labs have significant influence on the government, and IPO activities are creating tension. Anthropic has been dominating the cycle of releasing the world's best model for the last two or three releases.
Anthropic has the best talent, with most people wanting to work there. They are valued more than OpenAI. Anthropic was an underdog just a year or two ago. Their share of the enterprise API market is approximately 90%, while OpenAI holds about 10%. OpenAI has lost badly to Anthropic, and once Anthropic gained a lead with Mythos and Fable, catching up will be difficult.
The government is now stepping in to say play fair and slow down to give others a fair shot. The speaker gives a 45% chance that individuals in the US government want a piece of Anthropic's IPO action and are playing hardball to jeopardize the IPO, with some under-the-table financial arrangement eventually clearing things up.
Politico reported that the US government made up the claim that Dario was at a wellness retreat when they tried to reach him on Friday. Anthropic denies this, stating he was unavailable for about an hour, they provided someone else, and he got on the phone immediately. The speaker believes Anthropic on this point. This detail is viewed as part of the power and propaganda game.
The situation started with Amazon CEO Andy Jassy being the snitch. Jassy contacted the White House saying Fable is not safe, mentioning they asked it to find bugs in their codebase and it found some bugs. The White House actually reached out to Jassy for comment and advice. Jassy launched Fable, gave it his pet project, and when it found a bug, that was sufficient. By Friday morning, the issue reached the highest levels of the White House.
The speaker compares the situation to Game of Thrones and suggests a show called Game of AI, nominating Sam Altman to play Littlefinger as he plays all sides. Dario has been advocating for and begging for regulation, asking for a process requiring companies to put models through review, and previously stating he wants the government to have ability to take models off the market. Now that it has happened, he is saying not like this.
The speaker gives a 10% chance that everyone is colluding in some sophisticated power game. Mark Andreessen responded that 5D chess does not exist and it's all just people improvising. The speaker believes on a philosophical level that 5D chess is hard to plan ahead.
Another major possibility is that people are just figuring things out, doing stupid things, reacting emotionally, with someone being butt hurt about a missed call or ignored text message. The speaker compares watching VEEP versus The West Wing, noting that VEEP probably reflects reality better with idiots fumbling around trying to accomplish silly goals driven by pride and ego.
Before Anthropic released the model, the White House begged them not to release it. Anthropic previously refused the Pentagon unbounded access for autonomous weapons and spying programs a couple months ago, which the speaker views as prompting this slapback. Anthropic has too much power, and anything to suppress that power and put them in check will be done.
Shortly after the call, the Trump administration imposed export controls on Fable 5, restricting or banning access to foreign nationals. This means employees like Andrej Karpathy, who is allegedly not a US citizen, may not be able to access the model. The White House gave 90 minutes to take the model down with no details on the actual threat. The person said there was never any begging or asking to work together, just a declared 90-minute deadline.
Dario distinguishes between a specialized jailbreak and a universal jailbreak. The reported issue was a little edge case. Dario was shown the proof of concept and responded that this is what they are talking about. The speaker believes if it was a serious issue, Dario would take it down, but views this as a false flag.
When testing Fable, the speaker was not looking at code. When not examining code, users get really impressed by initial proof of concepts. The speaker asked Fable to make Factorio-type games in the web browser and it performed well. Non-trivial changes were requested across five or six rounds of rewrites to core game mechanics, and it did not stumble.
Executives get oneshotted by doing the same thing, taking a Friday or Saturday and throwing everything at it, getting impressed by the end result. This is not how to judge these models. Andrej Karpathy said when you look at the code, it gives a heart attack, but if you do not look, you save yourself the heart attack. For serious engineering creating products for millions of people, looking at the code is non-negotiable.
Nothing has really changed with code quality. LLMs struggle with code quality because they do not think long term, only solving the current problem. Some engineers also think short term, especially junior ones. The speaker prompted Fable nonstop from 8:00 a.m. to around 3:00 p.m. The model by itself does nothing and needs human guidance, direction, and quality assessment.
The speaker is not spooked about this replacing humans. Many companies are now changing their tune from replacement to augmentation. The cynical read is that backlash made companies change their tune, but the speaker thinks everyone is sobering up and recognizing it is a really cool calculator, an inanimate object with no sense of agency that needs direction.
Fable can work very long and is patient, taking its time, unlike previous models that rushed to finish. Opus 2 is mentioned as potentially similar. The speaker has not built something big like this for a couple of models, with the last deep session being around 4.6. ChatGPT 5 or GPT 5.5 is reportedly capable of much of the same.
The speaker does not see AI replacing people. Less people can do more, but the gap closes and companies can do more. Fundamentally this technology does not replace humans, it empowers them. This model is nothing without the human operator. A category error has been made on this technology.
There will be disruption, turbulence, and weirdness ahead. The speaker's stance is that everything is simultaneously true: it is autocomplete, but it turns out to be pretty damn useful. Nothing has really changed for the speaker personally. They hold reservations and are figuring things out, getting confused, lost, confident, and doubtful. Even people building this technology admit they do not really understand what will happen.
Keep Mo Bitar in your library
Save the videos and channels worth coming back to, and find them again in one place.
