nsblock

nakamoto-style blockchain / web3 🌐 / eth maxi

[Xatacrypt @xatacrypt]

here's what the Fable 5 release looked like until GPT-5.6 showed up

[Xatacrypt @xatacrypt]

I’m starting to like Anthropic more and more

Today I learned that the company is now trying to save the AI industry in the US

After Fable 5, Anthropic seriously started thinking about its future

> How can you release a new model if the US government can block it?

Anthropic, together with other tech companies, began working on a common system to evaluate the severity of jailbreaks

Fable 5 and then GPT-5.6 are advanced AI models that were quickly banned

Trump administration reacted fast to the jailbreaks found in these models

But how exactly did they assess the level of danger?

The government can simply take harsh measures against AI even if the problems are relatively limited

> Now Anthropic is trying to create a common scale for measuring the danger of AI vulnerabilities

The idea is to evaluate jailbreaks by several parameters:
- how powerful the capabilities it gives
- how easy it is to use
- how quickly it can become a real problem

This sounds like a good idea for solving the problem.
But it’s not as simple as it seems

> On one hand, it looks like an attempt to make the process more transparent

If there is a common evaluation system, reactions to found vulnerabilities may become more predictable in the future

> On the other hand, this system is unlikely to strongly limit the government

If necessary, it can still block a model by referring to national security

Also, the main question remains - who will find these serious problems first?

The AI companies themselves have an obvious conflict of interest

For me, the logical solution here is to create a government organization that will use Anthropic's scale and constantly check AI models for dangerous jailbreaks

Will the government go for this? I think it quite possibly will

AI is now too important for the US

These are my thoughts on this. What do you think?

[Xatacrypt @xatacrypt]

Why would Anthropic release a new Mythos model while Fable 5 is still has major issues?

Today on Polymarket, the chance that Anthropic will release the next Mythos-class model before August 31 is 37%

In my opinion, these odds are still too high

Fable 5 has returned, but it came with noticeable restrictions

> Right now Anthropic is mainly working on improving classifiers and reducing false positives. This takes time and resources

On top of that, pressure from the US government hasn't disappeared

In these conditions, launching a new full Mythos model before the end of August looks unlikely

Most likely, they will continue fixing the current version instead of rushing a new one

What do you think? Am I being too pessimistic about the chances?

[Xatacrypt @xatacrypt]

Claude Fable 5 is coming back - but at what cost?

Anthropic is restoring access to Fable 5 on July 2, but with significantly stronger restrictions

> It looks like the US government has won

The coding work in Fable 5 will now be done by Opus 4.8

The model that was strong specifically at programming will basically no longer be used for programming

> So what’s the point of bringing it back?

Most likely, Anthropic wants to recover at least part of its development costs and stay competitive in the frontier model race

> Because of the government's intervention, progress on the most advanced AI models will probably slow down noticeably

Whether this is good or bad in the long run is still hard to say

On one hand, controlling the most dangerous AI capabilities might be justified

On the other hand, it creates a precedent where the government can strongly influence the development of technology

What do you think? How justified is this kind of government intervention?

[Xatacrypt @xatacrypt]

While Fable 5 is still unavailable, Anthropic continues to improve its other models

Just an hour ago, Claude Sonnet 5 was released

What we know so far:
- it has become much more agentic
- major improvements in coding, reasoning, tool use, and knowledge work
- Sonnet 5 is now much closer in power to Opus 4.8, but remains cheaper

> Did the government review the new model before launch?

We don’t know for sure

But I think the company got approval from the Trump administration

I believe they will now monitor this much more strictly than before

They will only allow models that the government approves

For now, Anthropic is smartly continuing to develop the Sonnet line

> This helps them stay competitive in the AI race even while their strongest model remains restricted

These are my thoughts today. What do you think?

[Xatacrypt @xatacrypt]

Fable 5 come back soon?

There are growing rumors that the model could return to public access very soon

> Trump administration is close to removing the restrictions

It will most likely be available first only to users in the US

In any case, this is already clear progress in the negotiations

> It seems that Tom Brown understands much better how to communicate with the government than Dario Amodei did

The progress in the talks after he joined is noticeable right away

It will be interesting to see what the Pentagon and the National Security Agency say about the model

But Pete Hegseth, as we know, was quite negative about Fable 5 earlier

I'm just keeping an eye on the situation

Are you waiting for Claude Fable to return?

[Xatacrypt @xatacrypt]

GPT-5.6 will not be available for a long time yet

I think the full public access to OpenAI's new model will not be available until August

Polymarket currently gives an 89% chance of release before July 31

> In my opinion, this is too optimistic

Fable 5 has already been unavailable for about 2 weeks

And there is still no clear information when it will come back

> Now imagine that the "selected trusted clients" who got early access to GPT-5.6 find serious problems in the model

In that case, the release could be delayed even longer

> The only possibility for a quick release is if OpenAI prepared for the government review in advance

Then they could open the new model to the public fast

But right now that looks unlikely

What do you think?

[Xatacrypt @xatacrypt]

Am I living or just passing time?

Every update like this from @Polymarket reminds me we’re not just waiting

Small changes -> bigger things coming

Grateful to be here while the space is still being shaped

> Bullish on what’s next, whether it's L2 or $POLY first

What’s your take?

[Xatacrypt @xatacrypt]

US Government Takes Control of GPT-5.6

After what happened with Claude Fable 5, the government is now using a similar approach with OpenAI

> The Trump administration demanded that CEO Sam Altman release the new model in stages

Access to GPT-5.6 will not be given to everyone at once -> it will be provided only to selected clients after review

This is a new precedent in the history of AI regulation

Does this mean that now every new AI model release must be approved by the government? I think yes

What do you think about this? I believe we will now see new model releases much less often

[Xatacrypt @xatacrypt]

Claude Fable 5 lived for only 3 days

Did any of you manage to use it?

I was deeply focused on Chinese models and completely missed the release

Now I'm reading that in just those three days it impressed a lot of people

> How strong was it really? Was it worth it?

I would be interested to hear your thoughts

[Xatacrypt @xatacrypt]

The White House is tired of Dario Amodei

Is he to blame for the problems with Fable 5?

On Polymarket, the chances of Fable 5 returning have started to rise noticeably

Wired article gave us many details about Anthropic's talks with the US government

The most interesting part -> Anthropic replaced their main negotiator with the government

> They replaced Dario Amodei with Tom Brown, one of the co-founders of Anthropic

The Trump administration was simply tired of dealing directly with Dario Amodei

One government official even called him a "weirdo"

Now the negotiations are led by Tom Brown and Sarah Heck, Anthropic’s Head of Public Policy

After they took over, the dialogue improved a lot

> It seems strange to me that one of the best AI companies waited so long to change their negotiator, even though they clearly have people who are better suited for working with the government

It's possible that Dario's communication style partly delayed the process

> At the same time, I have no doubt that Dario is a technical genius. This doesn’t mean he is a bad person

Negotiating with politicians is a completely different skill that not everyone has

I'm glad we finally have some news. It will be interesting to see how Tom Brown handles this

What do you think about this?

[Xatacrypt @xatacrypt]

Top 3 Chinese AI models worth trying in 2026

A lot of people still sleep on them

GLM-5.2, DeepSeek V4, and Qwen3 are already very strong

A lot of people who tested them were surprised how good they are

Quick practical guide ↓

1/6

[Xatacrypt @xatacrypt]

Fable 5 is blocked - but China still won't become #1 by the end of 2026

Today I noticed that Polymarket traders give Chinese companies a 10% that one of them will take first place on the Arena Leaderboard by December 31, 2026

> Unfortunately, even these low odds seem a bit too high to me

As I said before, the problems with Fable 5 and the US government's intervention really create some opportunity for competitors

Anthropic now has to spend resources on compliance instead of quickly developing new models

In theory, this should help those trying to catch up

However, the main limiting factor for Chinese labs isn't algorithms or data

Their biggest problem remains the same - they have a severe shortage of advanced chips and computing power

> It is almost impossible for them to close the remaining gap and confidently take first place in the next six months

Even if Chinese models continue to improve quickly in coding and agentic tasks, scaling to the level needed to clearly beat everyone on a blind leaderboard requires a completely different infrastructure

Also, even if they had enough resources, they may simply not have enough time to build such a powerful model in just six months

> Recent releases from Zhipu (GLM) and Alibaba (Qwen) look strong

But going from "very competitive" to "the best model in the world" in such a short time seems unlikely

Overall, I like how well Chinese labs are progressing

> Zhipu AI's GLM-5.2 already shows excellent results in coding and frontend tasks, sometimes beating certain versions of Claude Opus in specialized benchmarks

> Alibaba's Qwen models also regularly rank high on many metrics

In the future, they could become serious competitors, especially if they manage to partially overcome hardware limitations through more efficient architectures and optimizations

These are my thoughts. What do you think about this?

[Xatacrypt @xatacrypt]

Anthropic may have reached the limit in frontier model development - and China could take advantage of this

It is clear that Anthropic’s latest model - Fable 5 - is significantly ahead of its competitors in logic, coding, and overall capabilities

> Let's assume they have already reached the point where further improvements bring naturally diminishing returns

And it is exactly at this moment that the US government's intervention makes special sense

If the model has become so powerful that its further development without strict limits carries serious risks, then regulatory pressure is a logical reaction

> The government is basically signaling: "go further -> dangerous zone"

As I wrote earlier, the speed at which the company solves its problems with the government will directly affect their success. I think the team understands this too

If Anthropic's progress slows down because of this, the entire competitive situation will change

> OpenAI, Google, xAI and others will get valuable time to catch up

In the race for the best AI models, they will definitely use this opportunity to close the gap with Anthropic

They can focus only on improving capabilities, while Anthropic will have to deal with regulatory costs.

> It's interesting that Polymarket still gives Anthropic high chances to stay the leader

But if my reasoning is correct, these probabilities may seriously underestimate the combined effect of hitting the development ceiling and regulatory pressure

> Even more important: Chinese AI labs are not standing still

Despite strict restrictions on chips and equipment, they continue to develop their own advanced models

The slowdown of the American leader may allow them to close the gap faster than expected.

As a result, the US government’s intervention - which is probably aimed at safety - may have the completely opposite effect

It may unintentionally speed up the global spread of powerful AI technologies

When the current AI leader faces external obstacles, America's dominance through unlimited scaling may come to an end

The main question is: are the US ready to accept such a weakening of their leadership?

They are not used to giving up positions on the international stage, especially in critical technologies

> Therefore, I admit that in case of a real threat to dominance, Washington may quickly make some concessions or soften the pressure

This, in turn, may create a dangerous precedent for the entire industry and other countries

It will be very interesting to watch how this develops. Every day, the future of AI becomes less and less predictable

These are my thoughts today. What do you think?

[Xatacrypt @xatacrypt]

Anthropic won't release next Mythos Model anytime soon

Today I noticed that Polymarket traders give a 69% chance that Anthropic will release a new Mythos-Class Model by September 30

> I think these odds are too high

On June 9, Anthropic released its first public Mythos-Class Model - Claude Fable 5

Usually, the time between their big model releases is several months

> But we shouldn't underestimate the company's current problems with the US government

Anthropic is spending a lot of resources to solve these issues

The longer this drags on, the more the next model may be delayed

> We also shouldn't forget that government restrictions on AI models - demanded directly by the US - is a new precedent in the history of AI development

It's hard to say right now what the consequences of this intervention will be

> But I'm sure this case has become an important example for all AI companies

How can they keep releasing more powerful models if the government can shut them down at any moment?

The constant race to create stronger and stronger AI may slow down a lot

These are my thoughts. I'd be interested to hear your opinion on this

[Xatacrypt @xatacrypt]

Claude Fable 5 will remain unavailable for a long time

Today I saw that Polymarket traders are giving a 70% chance that Fable 5 will become available again in the US

> I think these odds are clearly too high

The US government demanded that Anthropic completely ban access to the model for all foreign users

I believe the problem is much more serious than that

On June 15, Anthropic met with the US administration but couldn’t reach any agreement

As we know, Amazon also found a jailbreak in the model and reported it to the authorities.

> I think the US government is also demanding strong restrictions on Claude Fable 5 for all users

Fable 5 was positioned as a Mythos-class model with very strong coding capabilities

> If the company is forced to heavily limit the model, how groundbreaking will it stay?

After heavy safeguards, it could easily turn into "just another good coding model"

> Then what is the point of this "new" model?

These are my thoughts. I'd be interested to hear your opinions on this

[Xatacrypt @xatacrypt]

Claude Code creator:

"Auto mode was such a big step up - I just start Claude, it works, and I move on to the next task without reading most requests"

From basic coding assistance to autonomous agents handling bug fixes and code reviews

Designers and PMs now ship code too

Claude became the central hub of the workflow

In this reflection video Boris how development got way more creative and less tedious in just one year

Worth more than a $1000 vibe-coding course

[Xatacrypt @xatacrypt]

Claude Code + YouTube = $20k/month at $20 CPM? Wrong question entirely

Most creators are still wasting hours editing videos and arguing over image models

The ones actually cashing in do this:
- Tube Magic script + timestamps
- Claude Code + Higgsfield generates simple drawings on command
- Voiceover-synced 10-min videos in the highest CPM niche

No editing. No face on camera. No waste

High margins + affiliate upside. Low overhead

Everything else is cope

Full workflow 👇

[Xatacrypt @xatacrypt]

Boris Cherny talks a lot about CLAUDE​.md

But even he almost never mentions one of the best ways to save tokens in Claude Code

It's the .claude​/rules​/ folder and path-scoped rules

> A normal CLAUDE​.md file is loaded completely at the start of every session - even the rules you don’t need right now

Because of this, you waste a lot of extra tokens

In .claude​/rules​/, you can create rules that load only when they are actually needed

> You do this with path-scoped rules

Just add this block at the beginning of the rule file:

---
paths:
- "src/api/**"
- "src/backend/**"
---

After that, the rule will only load when Claude is working with files in those folders

> For example:
- you're working on frontend -> backend rules are not loaded
- you're working on API -> only the relevant rules are loaded

As a result, Claude gets only the context it needs and uses fewer tokens

Bookmark this if you use Claude Code

[Xatacrypt @xatacrypt]

Claude Code vs Codex? Wrong question entirely

They train completely opposite habits for working with AI agents

> Claude: steering agents up close through conversation

> Codex: dispatching agents in parallel with verifiable outputs, sandboxes and auto-review

No winner declared

The real skill in 2026 is agent literacy: knowing when to steer vs when to delegate - and always demanding proof before anything ships

Best operators use both strategically

Full video breakdown + practical Codex Skills in my article below 👇