Hazem

Hacking and AI

[Hazem @0xAxem]

Cursor might be a really good option for a subsidized Fable 5 tokens after it leaves the claude subscription

[Hazem @0xAxem]

@theo Pro tip, if you create a new account use the 20$ until you make then upgrade to 100$ then 200$. Usage reset after each upgrade

[Hazem @0xAxem]

I think AI labs should put more effort into making their models more aligned when it comes to implementation timelines. Fable just gave me a plan with one phase taking about 2 week and it took only one hour to complete using AI

[Hazem @0xAxem]

@TimsUndead @thsottiaux Tried it on two windows 11 machines, and still facing a lot of issues it hangs every couple of minutes

[Hazem @0xAxem]

@jxnlco Mostly search questions where I need the result to be shared across multiple devices.

[Hazem @0xAxem]

I have been doing this for a while, it gives you the best of of both worlds. if you want to try just do this:

1. install codex cli
2. inside of claude code run this

/plugin marketplace add openai/codex-plugin-cc
/plugin install codex@openai-codex
/reload-plugins
/codex:setup

3. Just add this to your claude.md

## Model Roles

- **Fable 5 (xHigh)**: Planner, Architect, Orchestrator, and Judge
- **GPT 5.5 (xhigh via Codex)**: Fast coder and executor

**Default approach**: Use Fable for planning, orchestration, and review. Use GPT 5.5 xhigh for most implementation work.

## Basic Rules

- Start complex tasks with Fable for planning and task breakdown.
- Delegate implementation to **GPT 5.5 xhigh** using the Codex plugin.
- Always review GPT 5.5’s output with Fable (never let it review its own work).
- Use Fable for architecture decisions, security logic, and final judgment.
- Use GPT 5.5 xhigh for coding, refactoring chunks, tests, and mechanical work.

## Codex Model configs

--model gpt-5.5 -c model_reasoning_effort=xhigh

[Mitchell Hashimoto @mitchellh]

I'm having a lot success using Fable xhigh as a planner/architect, using GPT 5.5 xhigh (subscription) as a coder, then Fable xhigh again as a judge. At API pricing, planning+judge costs are in the ~few dollar range compared to typical $50+ full round trips.

I've seen some others using dumber/cheaper coders, but GPT 5.5 even at xhigh compared to Fable 5 is very cheap and very fast. And GPT 5.5 is just... really good.

Still been less than 24hrs since the re-release so the longevity of this approach is unclear, but its been working really well.

[Hazem @0xAxem]

GLM 5.2 is scary good, I’m happy that this capability is open for everyone

[Hazem @0xAxem]

Why would OpenAI have a capacity issues if not because of a new model they are deploying 🤔

[Tibo @thsottiaux]

Oy. We are aware that some Codex users are experiencing high error rates with "model at capacity" and are working to bring things back to being stable.

https://t.co/R3dCKGGtQw

[Hazem @0xAxem]

What would Anthropic do if they were the inventors of gpt2

[Hazem @0xAxem]

@theo Feels a lot better than opus, but super slow to the point that I would still use gpt5.5 as my daily main model.

https://t.co/dmMJ3uIelI

[Hazem @0xAxem]

can't believe emojis everywhere is back with fable 5 🤣

[Hazem @0xAxem]

Fable 5 is not really usable for anyone outside anthropic or companies paying tens of millions of dollars to anthropic. It took more than 50 minutes to execute a simple plan.

GPT5.5 is way more usable (not necessarily better in benchmarks) than fable, I can iterate faster, and i’m not capped by only 8 prompts in a day.

I give the openai team 3 days to drop a model that is even better than 5.5 and a lot closer to mythos

[Hazem @0xAxem]

those questions are crazy, I guess the natural development for models is to have their own optimized language and a translation for humans.

[Hazem @0xAxem]

can't believe emojis everywhere is back with fable 5 🤣

[Hazem @0xAxem]

Good thing my claude code is using a 3 year old version of TanStack

[Socket @SocketSecurity]

🚨 BREAKING: 84 TanStack npm packages were compromised in an ongoing Mini Shai-Hulud supply chain attack, adding suspected CI credential-stealing malware.

Socket flagged every malicious version within six minutes of publication. This is a developing story.

[Hazem @0xAxem]

I think we should have more opinionated AI agents instead of generic agents

[Hazem @0xAxem]

clearly anthropic is not happy with me using claude in cybersecurity. I would love it if they have a feature verify my identity so I don't get banned

[Hazem @0xAxem]

I believe that this is the correct approach to deal with more advanced models or less restricted models. I believe we already reached the scary phase of vulnerability discovery after Opus 4.5 was released.

[Hazem @0xAxem]

I believe that this is the correct approach to deal with more advanced models or less restricted models. I believe we already reached the scary phase of vulnerability discovery after Opus 4.5 was released.

[OpenAI @OpenAI]

We’re expanding Trusted Access for Cyber with additional tiers for authenticated cybersecurity defenders.

Customers in the highest tiers can request access to GPT-5.4-Cyber, a version of GPT-5.4 fine-tuned for cybersecurity use cases, enabling more advanced defensive workflows.
https://t.co/RMMXQklFar

[Hazem @0xAxem]

I have been trying to max my $20 plan to upgrade to the $200 for 5 days now, I couldn’t get even close to the limits. Codex is winning the race

[OpenAI @OpenAI]

We’re updating our ChatGPT Pro and Plus subscriptions to better support the growing use of Codex.

We’re introducing a new $100/month Pro tier. This new tier offers 5x more Codex usage than Plus and is best for longer, high-effort Codex sessions.

In ChatGPT, this new Pro tier still offers access to all Pro features, including the exclusive Pro model and unlimited access to Instant and Thinking models.

To celebrate the launch, we’re increasing Codex usage for a limited time through May 31st so that Pro $100 subscribers get up to 10x usage of ChatGPT Plus on Codex to build your most ambitious ideas.