Find Your Signal, Find Your Subtraction

0:00 Sam Rogers

Signals and subtractions. What to watch, what to drop every week

0:00 Sam Rogers

Hey, welcome to Signals and Subtractions. This is episode two. I'm Sam Rogers. Uh, typically we'll have a co-host and a guest. Today it's just me. I wanted to set some ground rules here for how I think about what this show is and what it's not, and specifically, what do we mean when we say signal and subtraction around here? Um, by the end, you'll be able to find your very own signal and subtraction with or without the use of AI.

1:00 Sam Rogers

But first, because AI changes all the time, news. For the longest time, AI has just been a black box. Like we know that stuff is happening. We see the outputs, but we don't really see what's happening underneath.

1:00 Sam Rogers

We just know that we didn't design it or build it directly, and while we can't yet say that AI is conscious, we can peer into its subconscious now. Yesterday, Tuesday, July 7th, Anthropic put out a new paper and an open tool revealing the Jacobian Lens, as they're calling it, or JSpace. You'll find lots of YouTubers talking about this.

1:00 Sam Rogers

Some of them even read the paper. But I would encourage you to go and play with it directly yourself, just as I did last night, so yes, what I'm referring to is a global workspace in language models.

2:00 Sam Rogers

Anthropic put out this fancy little video, it's, it's actually very good, which compares what JSpace is to human consciousness. Um, there are links here to the paper, some of it's a bit beyond my depth, but definitely worth reading. Most fun of all, uh, you can go to their open tool, and you can actually see JSpace in action with two different, uh, large language models, uh, Qwen and Gemma.

2:00 Sam Rogers

And this is also a, a GitHub repo, by the way. So this isn't the thinking part of the model.

2:00 Sam Rogers

This isn't, uh, you know, in a reasoning model, sometimes you see the part at the top of it's thinking about something. This is before that. It's not the output tokens that you get. You're not being charged anything for JSpace. This is like a view into, um, what's happening underneath. So, um, what it reveals seems to be the precursors of all the responses that we see.

3:00 Sam Rogers

Um, that's the assertion of the paper anyway. Lots to talk about here, um, but not in this particular episode because we're only, like, 24 hours in. I'll put some links in the show notes. Why don't you put something from your JSpace in the comments right now? Let's just see what happens. So the reason that all of this is important, uh, the reason it's news is not that AI got smarter. It's for the first time we get to see inside this black box. It's a little transparent.

3:00 Sam Rogers

It's very fascinating stuff, and it's actually quite new. . Uh, Fable maxing. Fable 5 is currently the most powerful AI model that's commercially available. This is the one that was not commercially available for a couple weeks while the US government was deciding if we could have it or not. Um, pretty much everyone who's deep in AI has been very busy for the last week maxing out their access window before Anthropic's billing practices change.

4:00 Sam Rogers

Originally, that was supposed to happen yesterday. Now it happens on Sunday, July 12th. Yay. So if you haven't already hit your token limit, and you probably did, um, good news, there's more for a few days. Um, this is something that has been panned on X and, you know, there's lots of chatter out there about, "Oh, they nerfed it.

4:00 Sam Rogers

It's not as good as it was before," or, "It was always a hoax," or, "Does anybody think this isn't any good?" Um, my direct experience, this is by far the best model ever, and it's the first one that I've been able to use successfully for some massive tasks that I've had rolling around up here in the noggin for a while.

4:00 Sam Rogers

Um, if you don't know what to do with Fable in this window, just ask me, 'cause I'll tell ya. All right. The next thing already flew in there, uh, FrontierCo by Microsoft. So Microsoft launched a $2.5 billion organization. This is 6,000 people, um, industry experts, AI engineers who work either on site or i-in some way deeply embedded with customers.

5:00 Sam Rogers

And this is their strike team, their AI deployment unit that is focused on AI integration in enterprise workflows. I've been writing signals and subtractions for a few months now about how we're moving from pilots being like a 2025 thing into production 2026. So to me, this is just further evidence that, yeah, we're here now.

5:00 Sam Rogers

It's infrastructure time. Let's go. Uh, so that's some of the signal from out there in the wild. There's more stuff that's happening, plenty more stuff.

6:00 Sam Rogers

I've been writing "Signals and Subtractions" since June of last year. This show is an extension of that to make it more relevant, to offer different perspectives, make it more fun for me and for you hopefully as well. A lot of people can do news segments like I just did or tell people how they think the world should be, but I'm less interested in people talking about AI or writing about AI or giving all of their opinions about AI.

6:00 Sam Rogers

I'm much more interested in people who are actually AI-ing about AI. And for those of us who've been doing the work, as I have been for years, um, you know you can't keep up, right? Like, you can't pay attention to everything. All of this stuff moves at inhuman speeds, and we are merely human. So the signals that we're tracking become that much more important.

7:00 Sam Rogers

Also, there's a lot of things that used to make sense that just don't anymore, and we need to actively release and let those things go so that we can ascend up the exponential curve in terms of what we can produce and how we can coordinate, how we organize and work together. Basically, how we all navigate this change ourselves.

7:00 Sam Rogers

If you'd like to see examples of the signals and subtractions that I've produced for the last year, all of the newsletter stuff is archived on the site. What site you may ask? Why sigsub.show, of course.

7:00 Sam Rogers

Feel free to share on all the socials. That'd be awesome. You can sign up and be notified for everything. When I say signal, what I mean is like a personal shift in my own work.

7:00 Sam Rogers

It's something that's recent, it's something that's specific, it's something that's first person. So it's not a headline, it's not a trend, it's not a tool, although sometimes, you know, it's cool to share tools. Tools are super cool. Um, the test for you here, if you wanted to contribute a signal, would be that only you could say it because you saw it or you did it firsthand.

8:00 Sam Rogers

So AI agents getting smarter, um, that's, not a good signal. Just think, if a stranger could have said it on LinkedIn, it's probably not yours yet. Um, rather than telling you my signals off the bat, uh, I would like to show you how to come up with your own. Find your own signal and subtraction. There are some, AI copy-paste prompts.

8:00 Sam Rogers

We're gonna go through it the long way, so the, the way that's by hand. So first, I glance at my calendar. Uh, this is really just to jog my memory about who I've talked to, and remember like, oh, right, that meeting I had on Tuesday.

9:00 Sam Rogers

Like, there's... In the past week, I've had these different basically change management discussions. All this scope creep is landing on good people right now, and I, I have these conversations about, like, what to do about that. I may also have some notes in my Obsidian vault, you know, wherever you keep notes is great, about interactions and, and things like, um, like it's not the big project that wears you down, it's the recurring drag of all these obligations, the standard meetings, the weekly reports, the sync that you inherit and that just keeps running forever, um, because somebody had a need one time, and now there's a calendar commitment that sucks time from me and my team.

10:00 Sam Rogers

So if you're like me, probably nobody in the org chart sat you down and bestowed upon you this sacred duty. Uh, it just drifted onto your plate, and you kept doing it, and nobody ever noticed that it never really got assigned, right? It's just an expectation now. So all I wrote down in my notes was mandate drift.

10:00 Sam Rogers

But it reminds me about all of these meetings, ceremonies that, that only add and never subtract. Um, so, you know, I have that to work with. And notice that it passes the test because it's first person, right? Me. It happened this week. And only I could say that I had these particular conversations. So that's a good signal.

10:00 Sam Rogers

And the archive is full of all these, by the way. So signal done, onto the subtraction. A subtraction is something that you actually stopped, something you actually killed or refused notice it's past tense. It's made, not planned. And it doesn't start as advice for other people.

11:00 Sam Rogers

This may not be the most familiar way to think upfront, but that's kind of the point. If we're sloppy, our focus goes add, add, add, add until we just can't add anymore, right? We have to subtract in order to make room to grow, in order to lighten the load, in order to be able to ascend up that exponential curve.

11:00 Sam Rogers

There's only so many things we can take with us. So, um, do fewer meetings, not so good. Um, I'm thinking of fewer meetings. Eh, not a subtraction. Um, I killed our Monday status meeting and nothing broke. That's a good one because it cost you something and you did it anyway. So I think about the stuff that I stopped doing in the last week on the calendar.

12:00 Sam Rogers

I think of the recurring meetings that I did not schedule for Q3, and in this case, um, I did have to add one back. So far it seems I got away with dropping three and nobody said anything yet. Um, one of those actually I, I delegated two steps up the org chart where I don't think it will survive for long, uh, at least not as an hour meeting.

12:00 Sam Rogers

It might be a half an hour. But either way, it's not on me, so I win. I also think about the mandate drift conversations and the solicited advice that I presented on various forums. Um, in this case, uh, there was a particular one in a Slack group that I could go back and remind myself of. So cool. Here now I've got the foundations of both a signal and a subtraction.

13:00 Sam Rogers

Not bad, but not great yet either. So the final piece is to put on the analogy layer. The idea here is to make something familiar to the people, easily accessible. So Who here knows about coding? Anyone know about loops? They're all the rage these days. I'm looking for some way to make it tangible to the audience, right? To take something that they already know and transpose the thing that I'm talking about on top of that so that it feels good to them and they feel smart when they've done that.

13:00 Sam Rogers

So I could say, um, something about do while loops. A do while loop rechecks its exit condition on every pass. It runs until the goal is met, and then it stops, right? It's a loop, but It's a loop that writes its own ending. So by contrast, for loops run on a fixed count. So I could type fifty-two and it would run fifty-two times because the number fifty-two was set once at the start, and it was never questioned again.

14:00 Sam Rogers

Nobody asks at iteration twenty-eight whether this deserves to loop yet again. It's just pure count. So our recurring meetings are like for loops, and the really cruel part is, of course, there are for others. Uh, when we run a do loop, doing something for ourselves, we hold the exit condition. . So the subtraction here is not cancel all your boring meetings. That's kind of, you know, generic advice, hard to apply. Um, it's to eliminate for loop meetings.

14:00 Sam Rogers

It's to convert for loops into do loops, and I recheck this at the end of every quarter and adjust as necessary. Now, I've never thought about doing it this way. I'm just trying to think of the analogy of how it is to express it to other people. So now I can say a do loop knows when it's done. A for loop just knows how to count.

15:00 Sam Rogers

And by the end of the quarter, most of my calendar is usually just counting. I'm not lazy. Uh, it's just, it's way too easy to optimize for the wrong loop type, right? So I stop, I remove or convert, and I do this as a regular practice, which is true. Uh, so finally, the last piece is we wanna make this testable as a specific action that the audience can do.

15:00 Sam Rogers

So this isn't advice exactly. I might say something like, "Try this before your next week fills up. Open your calendar and look at your recurring events. For each one, ask two questions. One, what is its exit condition? Two, who checks it? If there is no exit condition or the person who checks it isn't there, you're running a for loop, and it's for somebody else.

16:00 Sam Rogers

Give that a return statement this week. No reason that you need to announce this or talk about it to other people. Just do it and try it." So that's a good subtraction. It's not advice or telling people what they should do. It's giving them something small and simple that they can try, and that's what we want in every episode of the future Signals and Subtractions. In that example, the signals and subtractions are two sides of the same coin like this. That's when it really works really well. So the signal's what you see, and then therefore you stopped, you know, the other side of the coin. But that's extra bonus points. Uh, don't try and force it. A true signal and a true subtraction will beat a clever pair of anything that's, you know, half invented.

17:00 Sam Rogers

I chose this example on purpose so that you could see it in action there we go. Do loop, and next we have The subtraction for loop. Good. See how that works? Okay, now you. If you'd like to take a copy of the home game, please visit sigsub.show/find-yours.

17:00 Sam Rogers

There's some clever AI prompts so that you and your bot friend can play whenever you'd like. And now the outro. Thank you so much.

17:00 Sam Rogers

Signals and

17:00 Sam Rogers

As part of my own Fablemaxing last Monday, I made an AI harness. Now granted, I've been thinking about these things for over a year, and I already had a lot of the raw materials lying around. The fact that I could do it in one day for version one was pretty fricking awesome.

17:00 Sam Rogers

Thank you, Fable.

17:00 Sam Rogers

Meet Harnessie, the ultimate brain-agnostic multi-agent harness. Connect any AI agent seamlessly and orchestrate them with absolute ease. Unlock the true power of your AI workforce