danieltanfh95

https://danieltan.weblog.lol/2025/06/agentic-ai-is-a-bubble-... We should be working to make HITL tools, not HOTL workflows where the humans are expected to just work with t…

Same. https://danieltan.weblog.lol/2025/06/agentic-ai-is-a-bubble-... The fundamental difference is we need HITL to reduce errors instead of HOTL which leads to the errors y…

Artifacts can reduce token usage by a factor of 2 or more. I had an [agent evolution framework](danieltan.weblog.lol/2025/06/agent-lineage-evolution-a-novel-framework-for-managing-llm-a…

Associativity is part of the contract/specifications of said "standard API"

a standard api is the magic. when you can talk to anything via a standard api it creates uniformity and standard blocks you can build from

These "talent wars" are overblown and a result of money having nowhere else to go. People are banking on AI and robotics for human progress to take off and that's just a result of all o…

The obsession with trying to explain a monad ultimately stems from conflicting explanations and the inability to differentiate between a mathematical monad and monads implemented in software. Monads i…

Its a harmful metaphor and clickbait title.

> Agent Lineage Evolution (ALE) is a novel framework that addresses the dual challenge of LLM agent degradation and manual prompt engineering bottlenecks. As LLMs inevitably degrade over extended u…

For reference this is the final prompt, though the initial meta prompt was included in the post https://danieltan.paste.lol/example-battler-data-final-promp... I am using claude sonne…

In this case I'm still converting to assets so I can ship asset files in future updates, but yes play testing, while doable on the editor, has to be recorded in code as well, which may not be as …

I already use ScriptableObjects with a custom inspector that outputs to YAML files. the problem was that with increasing use of nested generics+components I would need to spend alot of time customisin…

I basically gave it the prompt i mentioned in the blog post and it then generated a final prompt. the final prompt is fitted against the data i have but if you're interested this is the final pro…

Exploring the birth and death of LLM agents ala Halo and Ghost in the Shell

Are you referring to phishing attacks? How would they enumerate users through the random salt? Also, are you suggesting that they try to recreate the private key and enumerate through all possible val…

This is a demo of an authentication approach that combines the familiarity of passwords with the security of digital signatures. Instead of sending passwords to servers, users derive cryptographic key…

Human thought is not a solved problem. It is clear that humans can abandon conventional patterns and try a novel approach instead, which is not shown by our current implementation of LLMs.

This is consistent with AI usage patterns that people now internalise: start a new context everytime you have a new task. LLMs suck at dealing with context poisoning, intended or not, and the more inf…

Of course there is. Humans can pattern match as a means to save time. LLM pattern match as the only mode of communication and “thought”. Humans are also not as susceptible to context poisoning, unlike…

AI models are fundamentally trained on patterns from existing data - they learn to recognize and reproduce successful solution templates rather than derive solutions from foundational principles. When…