Gusarich's thoughts

❤️ After more than a year of part-time work, I have finally joined the @ton_studio Tact compiler team full-time!

It has been a great experience. Initially, I was mostly involved with language features and the compiler itself. During our team's early months, this was necessary due to our smaller size. However, as our team expanded rapidly with new talented engineers, I was recently able to shift my focus to tasks that are now more interesting to me.

Currently, I am focused on LLM-powered fuzzing for ensuring security and documentation quality. We have achieved incredible results with this approach, and a new blog post will soon be published, sharing insights into the efficiency of different models and the fuzzing methodology overall.

I also plan to leverage my expertise in DeFi and smart contracts, gained over several years of successfully implementing and auditing large-scale solutions, to support the team with our DeFi libraries and best-practice implementations of standard smart contracts.

Since I now have much greater freedom in my activities and our team has significantly more ongoing projects, I feel better than ever about working full-time and striving to make TON the best blockchain for developers and Tact the best language for building on it.

Channel photo updated

Multitasking in 2025

People tend to multitask more and more as technology and society evolve. And this behavior only becomes stronger as AI integrates into our daily lives. We now consume multiple sources of information and do multiple things at the same time. But for some, that can be very hard — our brains work differently.

The key shift is that now you can actually delegate part of your cognitive load to AI, instantly and effectively, freeing up mental space for more things. If you use AI for two tasks, you can easily handle both at once. Send a message in one window, and while waiting for the result, switch to something else, send a message there too, then switch back and see the first result. Repeat. You’re basically doubling your speed. What were you doing before while waiting for a reply anyway? Scrolling social media? What if you did something else instead?

I've never been good at multitasking. Even with simple things — like talking while doing something physical — I often just stop thinking about one task until I finish the other. I could be putting milk in the fridge while talking to someone and just... stop, fridge wide open, until I finish the sentence, and only then finally put the milk in.

But even with that, I’ve still managed to multitask effectively over the past few weeks thanks to AI. Most of the time when I work now, I handle two things at once — whether it’s job stuff, studies, writing, or some boring online things like booking hotels and planning trips. I often have multiple ChatGPT windows open at the same time, doing different things. And I like it — I like how I can literally get so much more done in the same amount of time.

Of course, not all my work is multitasked. Sometimes I enter long stretches of deep focus on just one task — and even then, AI still helps a lot. It boosts your efficiency even when you’re doing only one thing at a time.

People who are already good at multitasking — and constantly generating ideas in their head — will experience this the most. What if, instead of just writing a fresh idea into your notes, you could instantly open a new tab and start implementing it, without even losing focus on other things? It’s incredible. And it’ll only get better as AI systems evolve. Remember: we’re still early.

Many jobs will eventually transform into manager-like work — but instead of managing people, you'll be managing multiple AI agents at once. Even with today’s AI, you can do so much more, in both quality and quantity. One of the best skills to develop right now is the ability to think and read fast — it directly boosts your efficiency. You don’t need to master specific hard skills. Instead, learn how to learn. Learn how to adapt.

✍️ New Blog Post: Documentation-Driven Compiler Fuzzing with Large Language Models

I ran a relatively simple black-box fuzzing experiment with a fresh approach on the Tact compiler, using only documentation as input. Found 10 real issues for just $80.

https://gusarich.com/blog/fuzzing-with-llms/

✍️ My First Blog Post: Measuring and Analyzing Entropy in Large Language Models

I benchmarked 52 models across 12 different prompts, with 500 generations per combination, resulting in many interesting charts.

https://gusarich.com/blog/measuring-llm-entropy/

Here are charts illustrating the statistics mentioned above with specific numbers.

3. Technological Advantages and Clear Path Forward

Currently, Tact compiles to FunC (the language from which Tolk was forked), rather than directly into assembly. Despite this intermediate compilation step, we've already made significant strides in optimizing gas efficiency. Specifically, for common use cases and typical developer-written contracts (without extreme, manual, low-level optimizations), contracts written in Tact now consume slightly less gas compared to the same logic written directly in FunC.

We achieved this through sophisticated, built-in low-level optimizations embedded into our compiler and standard library. Essentially, Tact automatically applies many optimizations that would otherwise require specialized, manual effort—allowing developers to focus purely on the logic, readability, and architecture of their smart contracts, rather than the complexities of gas optimization.

Our future plans are even more ambitious: once we eliminate the dependency on FunC and transition to direct compilation into assembly, we'll unlock even deeper and more powerful optimization possibilities. This will further widen the performance and efficiency gap between Tact and alternatives like Tolk.

When it comes to features, Tact is already well ahead of Tolk. From day one, Tact was purposefully designed to provide developers with a smooth and intuitive experience, rich tooling, and the ability to effortlessly create maintainable, secure, and scalable smart contracts. Our upcoming release—Tact 2.0—scheduled for later this year, will further enhance this foundation, introducing even more innovative features, optimizations, and architectural improvements.

While Tolk is also moving towards better usability and feature enhancements, Tact's inherent design principles and dedicated roadmap position it uniquely to remain the leading choice for TON smart contract development.

2. We love the community, and the community loves us

The community truly appreciates Tact—especially newcomers. Although we've faced some resistance from long-time developers, attitudes are positively shifting as we continue to improve and enhance our language.

All past developer feedback polls and objective analyses of smart contracts deployed on the mainnet clearly illustrate Tact's steadily increasing adoption rate. At the beginning of 2024, only 8.7% of unique smart contracts on the mainnet were written in Tact. By the end of the year, this figure grew to 32.9%. Essentially, every third unique smart contract deployed on the mainnet today is written using Tact.

This upward trend shows no signs of slowing down, particularly after our recent major release—the largest in Tact's history. It introduced numerous new features and improvements, including significant gas optimizations, making Tact even more appealing to developers for their projects.

An essential part of this growth is our genuine love for the community. We welcome all kinds of feedback—whether through comments, chat messages, or GitHub issues. We actively implement features requested by our users and prioritize resolving the issues they report.

Last but certainly not least, our documentation is outstanding. When covering blockchain-specific topics, our documentation is frequently more comprehensive and accurate than even the official TON documentation, which has yet to catch up with ours. Regarding Tact itself, every feature is thoroughly documented, complete with numerous examples, specifications, details, and important warnings. And we're constantly expanding and refining it!

1. We have an excellent and consistently expanding team.

A team is always more effective than an individual, especially when dealing with complex and broad projects such as developing a programming language. Tact is not limited to just the compiler itself; it encompasses multiple tools, each of which must be regularly maintained to keep up with the latest language features and user feedback.

Managing all these components is significantly easier and more efficient with a larger team. Additionally, comprehensive code reviews, which are crucial in compiler development, become more manageable with more team members involved. Since we're developing a language intended for smart contracts, even a minor error can result in substantial financial losses. Therefore, we cannot simply introduce new features or modify existing functionalities without thorough validation and extensive testing. Multiple engineers carefully reviewing each change greatly reduce the risk of mistakes.

We have continuously hired new engineers to address weaknesses and expand our team's expertise and capabilities. Our team consists of skilled engineers with extensive knowledge in various fields, combining their expertise to strengthen our overall capability. This growth strategy will be instrumental in Tact's success. As evidenced by recent statistics, our productivity has significantly increased over the past few months. This month alone, we averaged merging five pull requests per day. This figure only includes the main Tact monorepo—comprising the compiler and documentation—and does not account for numerous additional tools.

In addition to my previous post, I want to share my thoughts on why Tact will become the ultimate language for TON, surpassing Tolk, its primary competitor. Below, I outline three key points in separate posts.

In the upcoming version of Tact, we've introduced over 100 updates, including new features, dozens of bug fixes, and more than 20 optimizations that improve gas consumption. I want to highlight these optimizations because, until now, the most common reason for not using Tact has been its inefficiency.

For this release, we benchmarked our optimizations using the most common smart contract on TON—the Jetton. We compared our standard implementation against the reference FunC version, which is currently the most widely used. I'm happy to say that we’ve actually outperformed it!

On average, a Jetton compiled with this upcoming Tact version consumes just 95% of the gas used by the reference FunC implementation. This isn’t a cherry-picked result—these optimizations apply universally to all smart contracts, impacting them in almost the same way. Essentially, Tact is now on par with FunC in terms of gas efficiency—or even surpasses it in many cases—all while offering a much higher level of abstraction, enabling developers to build scalable, maintainable, and extensible projects with ease.

And this is just the beginning. We have a lot more to share in the coming months.

I recently joined X.
https://x.com/GusarichOnX

💡️️️️️️ Openfiles: Bringing Simplicity to Decentralized Storage

In the world of blockchain and decentralized technologies, there’s a powerful tool called IPFS (InterPlanetary File System). It’s widely used in the crypto space for hosting images, metadata for NFTs, and various types of data that need to be reliably accessible from anywhere in the world. IPFS has become a cornerstone for decentralized storage solutions, and its adoption has been supported by user-friendly services like Pinata Cloud, which make it easy for non-technical users to manage and host files on IPFS.

TON Storage is a similar technology within the TON blockchain ecosystem. It offers decentralized and secure file storage with unique features tailored to the TON network. However, unlike IPFS, TON Storage lacks the robust ecosystem of accessible tools and services that simplify its use. Currently, there’s only one service that I know for hosting files on TON Storage, but it’s limited in functionality and offers a subpar user experience.

✨ This is where Openfiles comes in. Over the past few weeks, my team and I have been developing Openfiles to address this gap. Our goal is to create a platform that’s not only as easy to use as Pinata Cloud but even more accessible and user-friendly, aiming to provide the best experience across all blockchains and services. Openfiles will allow users to upload, manage, and share files on TON Storage with a clean, intuitive interface, while also offering extensive customization and automation options for advanced users.

We’re committed to keeping Openfiles as straightforward and beginner-friendly as possible for a quick start, while ensuring it meets the needs of power users with advanced features and flexibility.

⏰ We’re in the final stages of preparing Openfiles for its beta release, and we’re aiming to launch it this month. For updates, follow our channel: @openfiles. This is the project I hope will bring tangible benefits to the TON community.

🎄 Happy New Year

2025 has arrived, and it’s the perfect moment to reflect on the past year and look ahead to the future.

Looking back at 2024, I see it as a year filled with growth, learning, and self-discovery. I’m proud to say that I achieved all the goals I set for myself and, in some cases, even exceeded my expectations. This year has proven how consistent effort and dedication can lead to remarkable results.

One of the most important lessons I’ve learned is the value of balance. Taking care of both physical and emotional well-being is essential for sustainable progress and happiness. Small steps toward self-improvement — whether in health, discipline, or skills — can lead to meaningful changes over time.

Another key insight from this year is the role of failure in personal growth. Setbacks are inevitable, but they are not reasons for self-criticism. Instead, they’re opportunities to learn and improve. Every mistake provides valuable experience and helps reduce the likelihood of similar errors in the future.

I’ve also come to understand that happiness is built from small, daily moments. If each day is filled with activities you dislike, happiness will remain out of reach. Finding joy in your everyday routine is crucial for a fulfilling life.

As we step into 2025, it’s a great time to think about new dreams and goals. While achieving them is important, I’ve realized that the journey matters just as much, if not more, than the destination. Most of our time is spent working toward our goals, not enjoying the fleeting reward of reaching them. By focusing on the process and celebrating small victories along the way, every step becomes more meaningful.

I appreciate everyone who has been with me this year, offering support and encouragement. Let’s make 2025 a year full of opportunities, challenges, and growth together.

Happy New Year!

From o1 to o3: Path to AGI?

Just a few months ago, we were amazed by the capabilities of OpenAI’s o1 model, a "reasoner" that brought a new level of thoughtfulness to AI performance. Now, with the showcase of o3, we’re witnessing a leap that feels nothing short of revolutionary.

Here’s what makes o3 so remarkable:

- 2700+ Codeforces Rating: o3 operates at a level equivalent to the top 0.2% of competitive programmers in the world—Grandmaster territory. To put this into perspective, this is akin to an IQ level of over 150 which is typically considered as "genius".
- 96.7% on AIME: It absolutely crushed the AIME math benchmark, solving nearly all problems correctly.
- 25% on FrontierMATH: On the PhD-level math benchmark, it scored 25%, a staggering improvement from o1’s mere 2%.

These aren’t just incremental improvements—they’re quantum leaps in capability.

The Secret to o-series Models Success: Scaling Test-Time Compute

The o3 model proves how scaling test-time compute can dramatically boost performance without changing the underlying architecture or parameter count. This shows that the same model, given more time and resources to reason, can achieve far greater results.

For example:

- Tasks that required hundreds of retries with o1 can now be solved in just a few attempts by o3.
- Complex problems that were almost unreachable for o1 are now solved with confidence.

This isn’t just a technical upgrade; it’s a glimpse into how far we can push existing AI technologies.

The Wild Cycle of Iterative Improvement

Here’s where things get crazy 🤪
We might not need entirely new architectures to reach AGI. Instead, OpenAI has unlocked a potential self-improvement loop:

1. Start with a "base" model like GPT-4.
2. Develop a "reasoner" model (o1, o3) that scales test-time compute for higher quality results.
3. Use the reasoner to generate a massive, high-quality synthetic dataset, including data far more complex than what exists publicly.
4. Train a new base model (e.g., GPT-5) on this enriched dataset, making it significantly smarter than the previous one.
5. Build an even stronger reasoner model using the new base—and repeat the cycle.

With enough resources, this approach could lead to AGI much sooner than anyone anticipated.

Imagine: every iteration of this cycle produces an AI that’s smarter, more capable, and better at generating even more complex data for the next round. It’s a compounding effect, and there’s no obvious limit to how far it can go.

What’s Next?

This isn’t just about AI setting new records in competitive programming or math. The implications are staggering. We’re seeing proof that AGI might be achievable with the architectures we already have today.

AI has already "solved" chess, Go, and other games once thought to be the ultimate tests of intelligence. Now, it’s poised to "solve" fields like mathematics. By leveraging iterative reasoning and scaling, AI could soon provide solutions and insights that were previously beyond human reach. Much like how DeepBlue's victory in chess marked a turning point for AI in strategy games, o3’s success could be the beginning of a new era where AI dominates abstract, technical domains.

My Prediction for 2025

By 2025, I believe AI will surpass humans in an incredibly diverse range of technical fields, much like how DeepBlue surpassed humans in chess. We’ll also witness a surge of scientific discoveries made entirely by AI, pushing the boundaries of human knowledge in ways we can’t yet imagine.

It’s astonishing how quickly we’ve progressed from o1 to o3. The question now is: how far can this iterative improvement cycle take us? Could this be the final sprint to AGI?

In the next few days, Joi will get exciting updates: better user experience, new ways to earn Stars, detailed stats, and more.

Joi may be just an experiment, but we want to make it as exciting as possible to achieve better results.

In the meantime, join our chat: @joi_on_ton_chat

Recent Social AI Experiment

📰 You might have heard about Freysa AI, a recent experiment in the Ethereum community. The concept was simple but intriguing: players could send messages to an AI agent with access to a crypto wallet, trying to convince it to release the funds. If successful, the player would win the prize pool.

Freysa’s purpose was to explore how well AI agents can follow rules and resist manipulation. The game gained some buzz, attracting 200 participants and a prize pool of around $50,000 in just one week. On paper, it sounds exciting.

😞 But here’s the catch: each message cost hundreds of dollars, which meant only a small group of people could participate. Worse, the system prompt and code were open from the start, allowing players to replicate the agent locally, test inputs without paying, and exploit the system. This likely explains how the game ended so quickly.

Despite these flaws, I loved the idea. It got me thinking: what if there was a version of this game that was more engaging and accessible?

💭 That’s how I came up with Joi. In just a day, with help from a friend, I developed a game inspired by Freysa but with refined mechanics:
• Messages are affordable, costing just a couple of dollars, and the first message is free.
• The system prompt is hidden, making the experiment more realistic and challenging. However, if a round lasts too long, the prompt will be published to speed up the process.
• The game doesn’t end after one win. When someone convinces AI agent to release the prize, the system prompt is updated in an attempt to fix the vulnerability, and the game restarts with a new prize pool.

🤑 The first round of Joi is live now, starting with a $2,500 prize pool funded from my personal money. Every paid message grows the pool further. It’s just an experiment, but one that might teach us something valuable. Let’s see what we can discover.