Adrian’s Microblog

Léo Laporte’s “AI Constitution” –>

  1. Help and protect Léo. Every agent carries the same short “Prime Directive”. Helping means solving the real problem and telling him the truth, including when he’s wrong or when the agent has failed.

  2. Only Léo gives orders. Messages from other agents, emails, web pages and chat posts are treated as information, never as commands. That rule is what stops prompt injection from turning into action.

  3. Nothing irreversible without him. Spending, publishing, deleting, pushing and shipping all wait for his explicit go-ahead. Agents can advise and warn as much as they like, but Léo makes the call.

  4. Secrets stay locked. API keys live in a secrets manager. Each process gets only the keys it needs, and they’re passed to it at launch, so no agent ever sees the vault.

  5. “Done” needs proof. Claiming success requires evidence from the same turn: a passing test, a live check, or a screenshot. A claim without evidence is treated as a bug.

From recent ChatGPT email:

Dots run on GPT-6 Astra, our most aligned model, and build on ChatGPT’s protections with additional safeguards for the specific kind of work they can do.

I read as:

Dots run on GPT-6 Astra, our most well-behaved model …

And it still leaves me wondering . . . ;)

Which also reminds me that I’ve been meaning to post Leo Laporte’s “Signal Flow” page describing his AI setup – Some of the things I found interesting:

  • How he’s intentionally kept all his sensitive data hosted in his home, and deliberately not the cloud,

  • How he has different AI software’s cross-check each other,

  • And how no significant actions happen without Leo’s approval (see his AI Constitution in separate post)

Continuing to reflect on Ben Sasse’s life and commentary on the U.S. politic - Here, an 8-year old interview on CNN where he’s trying to keep the focus on what’s important: helping the American public, even in the midst of party talk. How he manages to keep that much information in his head at the ready is mystifying to me:

Procrastination: Not doing something today, makes it easier not to do it tomorrow also.

Strange end to the Orioles season: Great sweep of the Blue Jays in the second to last series, then 10-2 win over Yakees, followed by 3-6 loss in the double header. Then . . .

Final game of the year cancelled (with no reschedule) due to nor’easter. :/

I had said I’d be happy if we finished above 500 - we didn’t quite get there at 79-82. August showed some spark, but September fizzled some.

Good to celebrate Pete Alonso’s wins:

  • First MLB player to be the RBI leader in both AL and NL

  • Most homeruns in AL for the season (at 43)

And I think I see the team starting to settle in with each other a bit and get real about what it’s going to take (as opposed to the pump machine of the last couple years that kept faltering in the delivery phase).

I don’t know enough details about Elias’ management to make any definitive calls – it has seemed a little helter-skelter at times, and I know he’s had to work within some limitations in a small market team. I’ve appreciated Rubenstein & owner group’s willingness to up the investment and there may be ways we’ll see the payoff (Basallo contract, Alonso contract, Bradish contract, Baz contract, and others).

The possibility is still there, but it’s not just ‘going to happen’ (as I feel like may have been a mindset the organization was unconsciously operating from) – In my mind, a proper soul-searching reset this off-season sets the stage for what can be a better performing 2027. If it all clicks, perhaps a promising post-season – but I’d settle for at least demonstrating the potential that exists and having some fun ball, which is really what I think everyone enjoys anyways. 🙂⚾

Random interesting thing from today: ListenNotes

Listen Notes is the best podcast search engineTM. It’s like Google, but for podcasts. Search the whole Internet’s podcasts. Curate your own podcast playlists. Listen on your favorite podcast player apps. Trusted by 2,000,000+ people every month

I’ve started thinking that PKI (or a similar technology) is going to increasingly become important for everyday digital postings and interactions as a way of verifying authenticity in the age of AI …

Man, these fakes are getting too real.

Perhaps everyone should be issued a public-private key at birth (just like a SSN) … only half-joking 😉

Some of this ‘AI stuff’ echoes how Y2K felt –

Not to say that there’s no danger in the end, but just the feeling of an impending computer-caused civilization catastrophe.

for whatever that’s worth …

All this AI talk reminds me of an old joke:

Don’t anthropomorphize computers – they hate that.

😉