V roce 2025 se v AI kodovani ujal "deep modules" (simple interface, complex functionality) jako zasadni pattern. V podstate cele Unreleased je takhle napsane.
Rok 2026 ukaze, jak je to udrzitelne. Kod v modulech je v podstate "write once, edit never". Kdyz je neco potreba, cele se to prepisuje. A API kazdeho modulu hlidaji agenti, aby se pouzivalo a neduplikovala se funkcionalita.
Rikate si, ze takovy AI agentni system poroste exponencialne?
Ano.
Bude to zabava.
→ We Are Still Unable to Secure LLMs from #Malicious Inputs
schneier.com/blog/archives/202…
“This kind of thing should make everybody stop and really think before deploying any AI agents. We simply don’t know to defend against these attacks. We have zero agentic AI systems that are secure against these attacks.”
“It’s an existential problem that, near as I can tell, most people developing these technologies are just pretending isn’t there.”
#AI #LLMs #stop #agents #secure #attacks #problem
Nice indirect prompt injection attack: Bargury’s attack starts with a poisoned document, which is shared to a potential victim’s Google Drive. (Bargury says a victim could have also uploaded a compromised file to their own account.Bruce Schneier (Schneier on Security)
Lesson 5. The agent has to say yes, but SO DO YOU. Find *your* agent, not just any agent. Preserving your power to say no is important at many stages of the publishing journey (multiple offers, bad deal terms, etc.), and you want an agent who respects you and gives good advice. I can't stress this enough. 6/8
#WritingCommunity #agents #LiteraryAgent #publishing #editing #querying
Full post: sarahcypher.substack.com/p/fin…
Once your novel is done, here are five lessons it took me 20 years to learn.Sarah Cypher (The Bird's Eye)
