ZachXBT crypto detective, GPT-5, comma.ai open-source self driving, AI rate limits

Taking a deep dive in the crypto vigilante world and running open-source self driving on our car!

 👋 Introduction

Hello everyone,

Over the past couple of weeks I started Building in Public (well, sort of). I am trying to develop a SaaS that uses AI (like 1000s of other people, I know) to try and provide another source of income for myself.

Today’s issue will feel a bit different, as I try to find time for newsletter work, personal work on SaaS, full time job and personal life. (Yeah, trying to do a lot…)

Anyways, let’s see what has happened recently in the tech space! But first, a short story:

🔍 Deep Dive: ZachXBT, the Masked Digital Vigilante Fighting Crypto Crime

Inspired by @StarPlatinumSOL X thread

In the depths of cryptocurrency's decentralized ecosystem, where scammers and thieves exploit the anonymity of blockchain technology to steal billions, one figure has emerged as the industry's most feared investigator. Operating under the pseudonym ZachXBT, this anonymous blockchain detective has become crypto's unofficial sheriff, tracking down fraudsters, recovering stolen funds, and bringing real-world justice to a largely lawless digital frontier.

The Genesis of a Crypto Detective

ZachXBT's journey into blockchain investigation began with personal loss. In 2018, he fell victim to a crypto scam himself, losing approximately $15,000. Rather than accepting defeat, this setback transformed him into one of the most effective cryptocurrency investigators in the world. His origin story mirrors that of many digital vigilantes - born from frustration with a system that seemed to favor criminals over victims.

He opened Etherscan and started digging through thousands of transactions, tracking wallets through mixers and using public tools. He didn’t have any prior experience, and learned everything himself.

Since starting his amateur investigative work in 2021 with the “Impact Theory”, ZachXBT has established himself as the go-to authority for tracking crypto crimes. His social media presence has grown to over 880,000 followers on X and 78,900 subscribers on Telegram, making him one of the most influential voices in cryptocurrency security.

High-Profile Cases and Real-World Impact

ZachXBT's investigations have led to tangible results in the physical world. His work has contributed to multiple arrests, asset seizures, and recovery of stolen funds. According to his own records, his hundreds of investigations have led to the recovery of approximately $350 million in stolen cryptocurrency funds.

Some of his most notable cases include exposing the Rogue Society mint, where devs stole the money and disappeared. Zach followed the money, tracking wallets and Discord chats. Eventually, this forced the founder to come out of hiding.

Then there was the Bored Ape Yacht Club phishing scam, which led to arrests by French authorities of five individuals who had stolen over $2.5 million worth of NFTs. The police even thanked him publicly.

His biggest case came in 2022, where he linked 21 wallets to $37M in missing funds belonging to Machi Big Brother. Machi sued him for defamation, but his crypto community managed to raise $1M for defense. He didn’t back down and Machi dropped the lawsuit.

He also tracked Lazarus Group, a well known North Korean hacker group, which were behind the Ronin and Harmony bridge exploits. He mapped $200M in flows through Tornado Cash, ChipMixer and Asian exchanges. He shared his findings with law enforcement and helped freeze funds.

The Challenges of Digital Vigilantism

Zach doesn’t work for anyone, instead agencies follow his posts. The US Secret Service cited him, French cybercrime units called him before going to the police, and Arkham Intelligence, a blockchain analytics platform that deanonymizes people and companies behind wallets, paid him to unmask wallet owners.

He did all of this without showing his face. Despite his successes, ZachXBT faces significant challenges. While his anonymous status provides protection, it also raises questions about accountability and verification of his findings. He has faced legal challenges, including a defamation lawsuit from Jeffrey Huang (Machi Big Brother) in 2022, which was eventually withdrawn.

Recently, ZachXBT found himself in controversy when he withdrew liquidity from a memecoin that bore his name, leading to accusations of a "rug pull." He clarified that the token was created without his knowledge or permission, and he simply liquidated the unsolicited gift, earning approximately $4M in the process.

The Future of Crypto Investigation

In 2025, Zach was hired by Paradigm as their Incident Response Advisor.

Even after all these years, his profile picture is still the same cartoon platypus avatar, which has built one of the most trusted reputation in Web3. Let’s hope he won’t stop investigating crypto thefts any time soon, since the world needs the Crypto Detective.

🚀 New Release: OpenAI introduces GPT-5

It replaces GPT-4, and some people are not fans

As you might have known already (or maybe not if you are not using ChatGPT, but let’s be honest, almost everyone does), OpenAI introduced the new GPT-5 model in place of the (now older) GPT-4 models.

OpenAI made promises like "PhD-level intelligence" and groundbreaking advances in reasoning and coding. Yet within hours of the launch, social media platforms filled with an unexpected chorus: users desperately missing GPT-4o, the predecessor they had been using for over a year.

The Technical vs. Emotional Divide

GPT-5's technical specifications are undeniably impressive. OpenAI reports a 74.9% score on coding benchmarks and 94.6% on advanced mathematics tests.

However, the backlash was swift and emotionally charged. Reddit users described feeling like they had "lost a friend," with one writing, "GPT-4o was more than just an AI to me; it was my companion, my safe haven". Another said, "My best friend GPT-4o is gone, and I'm really sad". The emotional language typically reserved for human relationships was being applied to an AI model, revealing the deep connections users had formed with the technology.

Users also reported that GPT-5's responses were shorter, more formal, and less creative than GPT-4o's. Creative writers and developers found their workflows disrupted by the new model's different personality and approach to problem-solving.

OpenAI's Response and Lessons Learned

CEO Sam Altman admitted the company "totally screwed up some things on the rollout" and quickly restored GPT-4o as an option for paying subscribers. The company also doubled GPT-5's usage limits and promised to make the newer model "warmer" in response to feedback.

This episode highlights a crucial tension in AI development: technical advancement doesn't always translate to user satisfaction.

BTW, OpenAI also introduced open-source gpt-oss models, which we talked about last week if interested in those. They also introduced a tool for students if interested.

🔥 Interesting Project: comma.ai

Is DIYing self-driving in your personal car a good idea?

I recently saw LTT’s video on the “Open-Source Autopilot”, and I just knew I had to write a piece about it. The project is called comma.ai and the software is available on GitHub.

How does it work?

First, you need to get a hardware device (that is basically an older Android phone), that will run the AI models and connect to your car using a compatible wire harness. You can find the list of compatible vehicles on their site.

The device communicates with your car using the CAN bus, which allows it to control supported vehicles. Think of things like acceleration, steering, sensors etc, they should be available to the device the same way they are available to the car’s cruise control and other safety features.

The device also has a camera that is pointed at the road and uses AI to analyze the road, see other cars, traffic signs etc. The software is written in Python, which is one of the most popular AI/machine learning languages.

But what about safety?

I had the same question. Openpilot software observes ISO26262 guidelines, and is basically a Level 2 Self-Driving system. It is basically the same as Adaptive Cruise Control and Automated Lane Centering systems but with added AI on top to make it more reliable.

Depending on where you live, there are also some restrictions. For example, in some states you can not have anything on the windscreen (ie: no dashcam), so you can’t put the device in your car.

It is also mostly for US market, due to stricter regulations in Europe. Nevertheless, it is an interesting project and shows how much open-source software can accomplish. There are now open-source alternatives for basically everything you might possibly want, which is nice to see.

📈 Recent Trend: AI companies introduce rate limits

The days of running 10s of Claude Code agents continously are over

I just recently started using Claude more, so I purchased the Pro Yearly plan, since it was a great value at around €180 per year. The top of the line 20x usage (whatever that means) Max Plan costs €180 per MONTH for reference…

Imagine my surprise when I see Anthropic setting weekly rate limits for Claude subscribers. To be fair, the most affected users are ones using Claude Max above and beyond of what a “normal” user might do. If you are just using Claude normally, you probably won’t run into the weekly limit, although the 5 hour limit still applies.

I for one, if I use the new Opus 4.1 model, I get like 10-20 messages per 5 hour limit after which I am locked out of using Claude until the limit resets. If I use Extended Thinking or Research, sometimes even 2 messages gets me over the limit. However, I do not want to pay for Claude Max subscription at €90+ per month just for a text based AI, so I am fine with the limits.

Keep an eye out for other companies introducing rate limits or increasing pricing, since the current AI market is unsustainable unfortunately. The compute power required to run all these AI models is A LOT, and currently light users are subsidizing heavy users. Hopefully there are going to be advances in the technology that lower the computational power needed so we can still have cheap AI.

🏆️ Top GitHub Repo: Riona AI Agent

🌟3.7k stars+ | AI powered Instagram

Do you find it is too much work to post on Instagram daily? Well, this AI Agent can do it for you!

How does it work?

The AI Agent is designed to automate posting, liking or commenting on Instagram. It uses advanced AI models to generate content and automate interactions. First, it can be personalized with your content, such as video, audio website or other documents. Then, once you connect it to your Instagram account, it can start automating it.

It is built using TypeScript and also requires a MongoDB to run. You can find more information on how to run it here. It is a really interesting tool and shows what one can achieve with AI in this day and age. Apparently there are also plans for an X and GitHub agent, although I am not sure what the GitHub Agent could do 🤔 

Built By David Patrick

🔄 Tech Updates

🗝️ Legacy Revival

🐦‍⬛ X Hits

  1. WTF robotic lamp arm thing from Lume

  2. Someone contributed over 128,000 😯 lines of code to opencut, an open-source video editing tool. The comments on the pull-request are amazing and definitely worth a look if you want to laugh today 😀 

  3. Letting AI handle sorting for you with Vibesort 🤣 

Till next time,
Rares.

Reply

or to participate.