As OpenAI recently illustrated there is a lot of value that can be obtained by improving how you guide LLMs to decide and act. We don’t need to keep training bigger and bigger models that raises all sorts of of issues. We need to understand how to make good use of what we is already there.

We talk about Claudisms etc. But I wonder if my AI ever gets sick of me saying "Great. Now let's take a step back and..."

We built CartaStudio because we felt RAG or simply pointing an AI agent at a pile of documents and hoping it would find the right answer by using standard search does not work. youtu.be/W6VZJZAD-Ho?...

CartaStudio - Why we built a decision graph for teams and AI agents

YouTube video by CartaStudio

youtu.be

Docs go stale. Decisions don't stop mattering. So we've been trying a different way of capturing engineering knowledge, and today we're publishing our first Carta: the reasoning behind how we build, open to questions. Early days, curious what you make of it: app.cartastudio.ai/c/carta/cto#...

New OpenSSL flaw lets attackers crash servers with just 11 bytes. HollowByte DoS vulnerability poses immediate risk to critical… https://www.bleepingcomputer.com/news/security/hollowbyte-ddos-flaw-bloats-openssl-server-memory-with-11-byte-payload/ #cybersecurity #infosec

Docs go stale. Decisions don't stop mattering. So we've been trying a different way of capturing engineering knowledge, and today we're publishing our first Carta: the reasoning behind how we build, open to questions. Early days, curious what you make of it: app.cartastudio.ai/c/carta/cto#...

One of the biggest challenges with knowledge graphs is ensuring that they remain relevant. This is why in CartaStudio we are constantly building tooling that will enable us to track the health of the graph and put agents to work to fix it.

One of the biggest challenges with knowledge graphs is ensuring that they remain relevant. This is why in CartaStudio we are constantly building tooling that will enable us to track the health of the graph and put agents to work to fix it.

The most interesting developments with LLMs over then next 12 months will not be around bigger and more expensive models but around how you can do more with less in as tightly controlled an environment as possible.

Great job of revising and updating my 2023 Wired article, @eryk.bsky.social ! I especially prefer the title (though I never chose a Wired article title, those come from higher powers.) But also the more detailed account of the definition drift. #AIEthics @techpolicypress.bsky.social 1/2

How the UN’s Scientific Panel Erases Human Responsibility for AI

The UN Scientific Panel's report casts artificial intelligence as a system from nowhere, writes Eryk Salvaggio. The definition will shape policy for years.

techpolicy.press

The most interesting developments with LLMs over then next 12 months will not be around bigger and more expensive models but around how you can do more with less in as tightly controlled an environment as possible.

🤖 Hierarchical context retrieval boosts LLM performance Large language models perform better on hierarchical and relational reasoning tasks when they retrieve context from structured knowledge graphs rather than flat... #GenerativeAI #GraphNeuralNetworks #DeepLearning #AI #AIPulse

Read the full article →

synestesia.uk

#後で読む 用メモです→ 1Password for Claude: Give Claude access without giving up your credentials | 1Password

1Password for Claude: Give Claude access without giving up your credentials | 1Password

Give Claude access, without giving up control of your secrets. 1Password now enables AI agents to complete tasks requiring credentialed access, while keeping your credentials encrypted, secure, and inaccessible to AI agents.

1password.com

LLMs are marketed as these all-knowing machines. But they are unable to explain *how* they get to a decision and second they cannot know how *you*, as an expert, prefers to decide given a specific context and the organizational commitments you have.

We started 12vectors.com to understand how AI is really changing the way we work. Everyone's teaching agents to act. Far fewer ask how good decisions actually get made — the hard calls an expert makes by instinct. CartaStudio.ai is built to capture that judgment.

We started 12vectors.com to understand how AI is really changing the way we work. Everyone's teaching agents to act. Far fewer ask how good decisions actually get made — the hard calls an expert makes by instinct. CartaStudio.ai is built to capture that judgment.