Аналитика Telegram-канала
Подписчики
59
Охват за 24 ч
17
ER за 24 ч
28,81%
Публикаций
12
Динамика подписчиков
7 дней
30 дней
6 месяцев
Больше
16 ноя19 ноя22 ноя
Показаны последние 7 дней
Больше истории →
Охваты публикаций
13.1117.11
24
Средний охват
31
Лучший результат
Показаны 2 публикации — последние в канале
Больше статистики →
Упоминания
5
Упоминаний
2
Из них репостов
3
Каналов
0
Упоминаний за 90 дней
0
Из них репостов
0
Каналов за 90 дней
Последние посты
2 из 12 2 из 12
How to Build an Agent
🎓 luma.com/zctjp8tt
In junior-level classes, we examine article How to Build an Agent by Thorsten Ball, which clearly demonstrates that there is no hidden magic behind AI coding agents, he provides a step-by-step guide to building a fully functional CLI-based AI agent in under 400 lines of Go code simply by combining a message-processing loop with basic filesystem tools
Even if you have never written code in Go, that is no obstacle; with a bit of effort, you can grasp the fundamental concepts of the language in a single day, if you already have experience with another programming language, you could even adapt the logic described in the article to a language you know, if you are a complete beginner, it might take a little longer but I am confident that, with enough dedication, you could work through the material in a week, even starting from scratch, by spending just an hour a day on it
Understanding these 400 lines will provide key insight into what autonomous agents which you likely interact with every single day actually are, these agents come in various forms: web-based chat interfaces (like ChatGPT), AI-powered chat tools integrated into IDEs, dedicated desktop applications, and CLI-based agents. Collectively, these are often referred to as agentic harnesses, the most important takeaway is that these are, at their core, ordinary programs built using conventional logic
What makes them artificial intelligence is the way they use APIs to send specific requests to an LLM which then are hosted in large data centers, the article explains this interaction in detail
During our sessions, we realized that the article is specifically designed for use with an Anthropic API Key, which could potentially be a stumbling block, i considered adapting the material to allow for other options, but then I realized that isn’t necessary in fact, navigating these challenges might actually provide you with some valuable extra knowledge
Here are a few things to look into and understand:
🌕Find out what ANTHROPIC_BASE_URL is
🌕If you use a custom ANTHROPIC_BASE_URL, the provider might not offer the specific model mentioned in the article, you’ll need to learn how to view the list of available models, select an appropriate one, and update the code accordingly
🌕Alternatively, you could explore how to adapt the code to work with OPENAI_BASE_URL and OPENAI_API_KEY
I think figuring this out will be useful for everyone, and you’ll definitely face various challenges along the way, by overcoming them, you’ll gain valuable knowledge and experience
The first challenge you’ll face is figuring out how to get the correct values for *_BASE_URL and *_API_KEY, as well as understanding concepts like ANTHROPIC-compatible and OPENAI-compatible protocols, there are many LLM providers that support these protocols and offer a free tier for example, check out qwencloud.com or dahl.global, additionally, the OmniRoute project catalog lists options that provide roughly ~10B tokens per month and openrouter.ai provide some model for free
As a result, once you understand these 400 lines of code, you will stop perceiving AI as magic and begin to see it as a tool for bringing your ideas to life
#note
ещё
31
1
17 ноя, 22:41
📁 Harness²: Recursive Agent Harnessing for an Open World
LLM agents’ capabilities depend not only on their underlying models but also on their execution harnesses. Recursive self-improvement (RSI) offers a promising approach to automatically evolve these harnesses. However, deploying RSI in open-world settings still faces two critical challenges. First, open-world tasks are typically unseen during the harness design phase and novel at test time, leaving no ground-truth verifiers to reliably guide online harness edits. Second, the RSI process remains highly task-specific and fails to generalize to novel scenarios. We introduce Harness², a plug-and-play, recursive agent harnessing framework that addresses these challenges through two connected levels of recursion. Tasklevel recursion refines the current task’s harness using a contrastive Propose–Probe–Compose step that reliably guides edits without ground-truth verifiers, and can be scaled further via parallel or sequential recursion.
🐱 github.com/google-research/harness2
📁 The avalanche that hit math is coming for AI development
- Main Idea: Artificial intelligence has already revolutionized mathematics and programming, and the same exponential wave of progress is expected to spread to AI development itself
- What Is RSI: Recursive Self-Improvement is a process in which AI researchers (autonomous agents) begin writing code, designing architectures, and training subsequent, even more intelligent versions of themselves
- Prediction: According to the author, automated AI systems capable of replacing the world's leading machine learning researchers and initiating this cycle could emerge as early as 2027-2028, dramatically accelerating the arrival of the technological singularity and superintelligence by 2030
#llm #rsi #harness
ещё
17
0
13 ноя, 19:27
Продолжить анализ этого канала
- История роста подписчиков
- Просмотры и охваты публикаций
- Кто упоминает канал и кого упоминает он
- Похожие каналы и подбор для рекламы
Открыть аналитику
Демо до 14 дней