. The agent deciding to act and the agent being able to act safely are two different problems.
Absolutely true
17 Matching Annotations
- Last 7 days
-
friends.of.trevorwcurr.ie friends.of.trevorwcurr.ie
-
-
Hindsight, an open-source memory architecture out of Vectorize, is a useful look at what a real answer looks like: instead of treating memory as raw logs you semantically search over, it separates facts, experiences, summaries and beliefs into distinct structures, time-stamps and links them to the entities they belong to, and lets the system hold a belief with a confidence score that updates as new evidence comes in rather than just appending the newest version on top of the oldest.
This is interesting.
-
Email me if you have questions, thoughts or comments! agnes@firstcircle.capital
General thoughts - this could be like 4 articles! It's depth and breadth are a lot to cover here.
-
good governance
If not executed well, it can destroy companies.
-
becomes "what is it allowed to touch?"
😂
-
Agent lock-in comes from accumulated context
This is something that businesses will need to start to think about. AWS AgentCore does have some interesting concepts around memory/context separation: https://aws.amazon.com/bedrock/agentcore/resources/?trk=9cd2d6e3-79cb-48fa-b04c-0233ea17f395&sc_channel=ps
-
AI as software but AI as a colleague
Mostly agree. AI as a 'colleague' is strong, but I agree with the sentiment.
-
(Read the paper on how the transformer revolution started "Attention Is All You Need" (Vaswani et al., 2017).)
Do you have a link?
-
retrieval-augmented generation: grounding a model's answer in retrieved data rather than its own training; more on this in the database orchestration section below
Maybe provide a link? this is a mouthful
-
high risk environment such as financial
100%
-
I got to test this becoming useful through tools
Perhaps
I tested this through
... oops you don't want style
-
e not reliable?
Ah, perhaps you mean reliable response performance. I was thinking of software reliability ("does it work") rather than performance consistency ("it works this way")
-
unlocking maximum productivity with agents reliability is going to be a core component.
I don't follow the reliability word here; but I do agree with the sentiment of 'unlocking maximum productivity with agents will be the core component" - i feel efficiency, safety will be elevated. Reliability is important, but people are more worried about cost and security right now.
-
. The human is now no longer the operator but the supervisor.
Agree!
-
robotic process automation bots
Probably, just
process automations
-
brilliant friend
brilliant, huh?
-
-
projects.trevorwcurr.ie projects.trevorwcurr.ie
-
Fine Arts
Should introduce the Siam Reap workshop that has demonstrations of high-quality craftsmanships.
-