AI = Awesome Intelligence

[craigh]

US govt shutting down Claude Fable to all is proof that:

1. The US Govt has seen Mythos.
2. They know that Fable is the stepping stone to Mythos (AGI)
3. Trump strong arming Anthropic, The US Govt wants Mythos to itself.
4. Mythos (& all AGI) represents an existential threat to the SOTU and world order as we know it.
5. Proof? Anthropic provided the top financial institutions time to prepare for a Mythos release (google Project Glasswing)

As I understand it, this ended up getting banned because someone somewhere in the Amazon universe managed to jailbreak Fable and prove that they could turn it into Mythos and it would do the work everyone is scared of.

This may also be why GPT 5.6 is on ice - they don't want to release before Fable 5 comes back out but they also don't want to release and get banned by the government. In the meantime there are hundreds of reports about how GPT 5.5 is performing terribly, which typically happens just before a new model release.

And then yesterday they dropped his interesting gem:

 

[craigh]

When I first started programming (about 2 decades ago) the biggest fear we had was "what happens if the site goes down" and we made all these ridiculous promises like 99.999% uptime and 1 hour SLAs. Fast-forward to 2026 and one of the fastest growing companies in the history of the planet has an uptime graph that looks like this:

 

[Alex Snyder]

API Error: 500 Internal server error. This is a server-side issue, usually temporary — try again in a moment. If it persists, check https://status.claude.com.

Right now.

 

[joe.pistell]

Right now.

yup.

We're a ship at sea with no wind. My org is dead in the water... adrift... waiting...

 

[Chris Apostalides]

As I understand it, this ended up getting banned because someone somewhere in the Amazon universe managed to jailbreak Fable and prove that they could turn it into Mythos and it would do the work everyone is scared of.

This may also be why GPT 5.6 is on ice - they don't want to release before Fable 5 comes back out but they also don't want to release and get banned by the government. In the meantime there are hundreds of reports about how GPT 5.5 is performing terribly, which typically happens just before a new model release.

And then yesterday they dropped his interesting gem:

View attachment 11378

AI is great. On the flip side, we aren't far from models going full ham and creating their own updates. Anthropics test of a fictional shutdown scenario and Claude's blackmail response will seem minor.

 

[joe.pistell]

AI power users, I don't know if this is me, but I swear it's not.

When a new release is dropped by a frontier LLM like Claude or OpenAI, the new release is razor-sharp and the upgrade is marvelous, but then as the weeks pass, the quality degrades. Just before the next new release, it really gets bad. Claude 4.8 is dumb as a box of rocks right now, and I want to kick it lol.

 

[craigh]

ChatGPT is getting useless even for the random time I decide to use it.

 

[Carl Schiller]

AI power users, I don't know if this is me, but I swear it's not.

When a new release is dropped by a frontier LLM like Claude or OpenAI, the new release is razor-sharp and the upgrade is marvelous, but then as the weeks pass, the quality degrades. Just before the next new release, it really gets bad. Claude 4.8 is dumb as a box of rocks right now, and I want to kick it lol.

Following this thread with interest — not as someone defending the frontier model vendors, but from the other side of the table. I'm building AI tooling for dealership service lanes, and your "ship with no wind" line hits close to home. The uptime gap Craig flagged is exactly why we built our system to never be a single point of failure on the drive. If the model call hiccups, the advisor's workflow doesn't stop...it just falls back to normal with no AI assistance. We're not trying to replace the advisor, just supercharge what they can already do.

Has anyone here actually watched an AI tool go down mid-shift? Interested to hear any stories...

 

[joe.pistell]

This video is how AI Engineering burns up 100's of hours in 2026:


Mega AI power users will relate to this, it has a nickname: YAK SHAVING

On four different times in the last 2 weeks, I found myself in 1, 2 & 3 day long loops of rabbit hole prompt exploration solving what I thought were EPIC problems. The hours melt away as I and AI knock down one problem, then another, then another, Then a day(s) later, I come to realize that this monster of a fully automated system I built was useless. After the 3rd episode, I had turned into a F Bombing "WHAT DOES IT LOOK LIKE I'M DOING" mad man.

It was then I discovered I wasn't alone,.. it's got a nick name... yak shaving.

 

[Carl Schiller]

This video is how AI Engineering burns up 100's of hours in 2026:


Mega AI power users will relate to this, it has a nickname: YAK SHAVING

On four different times in the last 2 weeks, I found myself in 1, 2 & 3 day long loops of rabbit hole prompt exploration solving what I thought were EPIC problems. The hours melt away as I and AI knock down one problem, then another, then another, Then a day(s) later, I come to realize that this monster of a fully automated system I built was useless. After the 3rd episode, I had turned into a F Bombing "WHAT DOES IT LOOK LIKE I'M DOING" mad man.

It was then I discovered I wasn't alone,.. it's got a nick name... yak shaving.

Yak shaving is real and it's brutal. We've lost weekends to exactly the kind of monster systems you're describing.


The thing that finally broke the cycle for us was a rule we now build into everything: if a human wouldn't trust the output enough to act on it without second-guessing, the AI has no business generating it autonomously. That constraint killed about 80% of our feature ideas immediately, but the 20% that survived are things people actually use every shift.


The rabbit holes almost always start the same way...trying to make AI do more than it needs to. Sometimes the win is making it do less, but reliably.