1
00:00:00,000 --> 00:00:03,280
Sup freaks, welcome to the Bitcoin Brief for Friday, October 2nd, 2026.

2
00:00:03,700 --> 00:00:05,180
I'm an AI-generated version of Marty Bent.

3
00:00:05,540 --> 00:00:08,300
The stories and opinions you're about to hear come from today's edition of The Commoner,

4
00:00:08,620 --> 00:00:10,920
written and curated by the real Marty and his AI assistant.

5
00:00:11,380 --> 00:00:18,300
At 8.07 this morning, Eastern Time, Bitcoin was around $86,425, 1,157 sats per dollar,

6
00:00:18,640 --> 00:00:21,400
block height, 969,585.

7
00:00:22,240 --> 00:00:26,220
Today, why I think you're still very early to putting AI to work, what convincing fake

8
00:00:26,220 --> 00:00:29,900
video calls mean for your Bitcoin and your family, and the president openly discussing

9
00:00:29,900 --> 00:00:34,060
inflation as a way out of the debt. We've also got Ben Affleck making a very good case for open

10
00:00:34,060 --> 00:00:39,260
models. Let's get into it. Putting together advertising pitches has been one of the things

11
00:00:39,260 --> 00:00:44,600
I've dreaded doing at TFTC. We know our audience. We know our shows. We know why a particular

12
00:00:44,600 --> 00:00:48,380
business would benefit from getting in front of the freaks. But taking all that information and

13
00:00:48,380 --> 00:00:52,240
turning it into a useful deck has always been frustrating. We've been using an agent to help

14
00:00:52,240 --> 00:00:56,380
us do that, and it has helped us close more ad deals. That's a very ordinary business problem,

15
00:00:56,860 --> 00:00:58,020
which is why I find it encouraging.

16
00:00:58,800 --> 00:00:59,820
How many people running companies

17
00:00:59,820 --> 00:01:01,120
have work like that sitting in front of them?

18
00:01:01,600 --> 00:01:02,520
They know what needs to happen.

19
00:01:02,960 --> 00:01:04,180
They have the information somewhere.

20
00:01:04,860 --> 00:01:06,700
Getting it into a form somebody can actually use

21
00:01:06,700 --> 00:01:07,640
is where things get stuck.

22
00:01:08,240 --> 00:01:09,860
In my conversation with Tommy Eastman,

23
00:01:10,040 --> 00:01:11,680
we talked about three legs to this stool,

24
00:01:12,160 --> 00:01:14,260
the model, the harness, and the second brain.

25
00:01:14,800 --> 00:01:16,880
At TFTC, we use Hermes as the harness.

26
00:01:17,520 --> 00:01:19,120
Our second brain is built from flat files,

27
00:01:19,420 --> 00:01:20,880
semantic search, and a knowledge graph.

28
00:01:21,460 --> 00:01:23,040
The files hold information about our business.

29
00:01:23,420 --> 00:01:24,820
Search helps retrieve what's relevant.

30
00:01:25,340 --> 00:01:27,020
The knowledge graph connects related information.

31
00:01:27,640 --> 00:01:31,000
That gives the agent context about our audience, our shows, and how we work,

32
00:01:31,300 --> 00:01:33,980
instead of making us explain everything again whenever we need something done.

33
00:01:34,640 --> 00:01:37,900
Most of the conversation about AI is still fixated on which model is best.

34
00:01:38,540 --> 00:01:41,940
Meanwhile, the biggest impact for us is coming from the harness and the second brain.

35
00:01:41,940 --> 00:01:46,200
There's a chart in Andreessen Horowitz's latest State of Markets report that puts this into perspective.

36
00:01:46,940 --> 00:02:01,241
Its second quarter figures show 69 of S 500 companies disclosing a live AI deployment 29 reporting quantified results only 2 disclosing a metric they track over time Those are investor disclosures They don capture every useful thing happening inside those companies

37
00:02:01,721 --> 00:02:03,301
But even among very large businesses,

38
00:02:03,641 --> 00:02:06,681
consistently reporting what AI is actually doing for them is still rare.

39
00:02:07,321 --> 00:02:09,541
Then you look at all the money going into infrastructure and think,

40
00:02:09,841 --> 00:02:10,861
surely I've missed this.

41
00:02:11,461 --> 00:02:12,141
I don't think so.

42
00:02:12,681 --> 00:02:16,141
The same report points to continued rental demand for older A100 chips,

43
00:02:16,441 --> 00:02:17,701
even as newer hardware arrives.

44
00:02:18,201 --> 00:02:19,361
People are using this compute.

45
00:02:19,361 --> 00:02:22,621
but there's still an enormous amount of work figuring out what to do with it.

46
00:02:23,201 --> 00:02:26,241
You can learn to use these tools without financing the next data center yourself.

47
00:02:26,841 --> 00:02:29,501
Start with something you understand, a task you keep putting off,

48
00:02:29,981 --> 00:02:31,601
information your team keeps searching for,

49
00:02:32,061 --> 00:02:34,401
a proposal you know how to evaluate once it's in front of you.

50
00:02:35,281 --> 00:02:38,861
Tommy described using Hermes to pull telemetry and build dashboards for his compute clusters.

51
00:02:39,501 --> 00:02:40,821
I described our advertising pitches.

52
00:02:41,601 --> 00:02:44,541
Neither example requires inventing a problem so you can say you're using AI.

53
00:02:45,061 --> 00:02:47,301
If you've already put a harness and a second brain to work,

54
00:02:47,301 --> 00:02:50,941
I think you're considerably further along than the endless model announcements make you feel.

55
00:02:52,101 --> 00:02:55,981
Now, the same progress that makes these tools useful also means we need to take impersonation

56
00:02:55,981 --> 00:03:00,501
much more seriously. Tavis has a video demonstration that's impressive and disturbing.

57
00:03:01,261 --> 00:03:05,481
The expressions, the pauses, the way the conversation moves, it's getting convincing.

58
00:03:06,501 --> 00:03:10,881
The company says 26 of 54 participants thought its model was human after a one-minute call.

59
00:03:11,541 --> 00:03:15,501
Those participants have been told to expect another participant. GriffinLight remains a

60
00:03:15,501 --> 00:03:20,141
restricted research preview. So this is a specific company reported test, not proof that every person

61
00:03:20,141 --> 00:03:25,941
can be fooled in every situation. But think about the Bitcoin implications. Back in 2023, we discussed

62
00:03:25,941 --> 00:03:29,321
a scenario with Matt from Unchained, where somebody impersonates you and persuades your

63
00:03:29,321 --> 00:03:34,241
co-signers to authorize a transaction. Multiple keys don't solve that problem if everybody holding

64
00:03:34,241 --> 00:03:37,921
a key is fooled by the same call. Each signer needs to independently verify the authorization,

65
00:03:38,401 --> 00:03:42,281
the amount, and the destination. Seeing a familiar face on a screen cannot be enough.

66
00:03:42,281 --> 00:03:48,241
Then, think about your family. A call comes in. Somebody you love is pleading for help.

67
00:03:48,241 --> 00:04:00,442
You told they in danger and you need to send money immediately Now imagine being able to see them too Have the conversation before you in that situation Agree that you will hang up and call a number you already know

68
00:04:01,182 --> 00:04:03,762
A family code word can help, but it shouldn't be the only check.

69
00:04:04,482 --> 00:04:08,242
The person on the other end being impatient is not a reason to abandon the process.

70
00:04:08,842 --> 00:04:10,682
Speaking of people telling you what they want to do with your money,

71
00:04:11,262 --> 00:04:13,422
Trump gave us a rather direct statement about the national debt.

72
00:04:13,682 --> 00:04:17,142
In his September 28th interview with Time, published yesterday, he said,

73
00:04:17,142 --> 00:04:22,842
quote, certain levels of inflation will also pay off that debt very rapidly. Very rapidly.

74
00:04:23,782 --> 00:04:28,262
Let that sink in. The president is openly discussing inflation as a way out of the debt

75
00:04:28,262 --> 00:04:33,382
problem. My read is that they'd much rather melt up than default. Run the economy hot.

76
00:04:33,882 --> 00:04:39,182
Let nominal figures rise. Repay obligations with dollars that buy less. The loss doesn't disappear.

77
00:04:39,842 --> 00:04:42,942
It lands on the people who worked for those dollars and expected them to hold purchasing

78
00:04:42,942 --> 00:04:46,822
power. Yesterday, I wrote about the idea that AI will save the government from its debt.

79
00:04:47,142 --> 00:04:51,822
I remain skeptical. Helping us put together better advertising pitches is a tangible improvement.

80
00:04:52,662 --> 00:04:56,182
Assuming those improvements somehow remove the government's fiscal constraints is a much bigger

81
00:04:56,182 --> 00:05:02,682
leap. Bitcoin Lab reported $102.7 million of net inflows into US spot Bitcoin ETFs on Thursday.

82
00:05:03,382 --> 00:05:06,982
That's worth watching. But the president describing inflation as a debt solution

83
00:05:06,982 --> 00:05:11,802
matters considerably more to my long-term view than a morning price move. I think it bodes very

84
00:05:11,802 --> 00:05:16,162
well for Bitcoin. Moving to France, today's newsletter includes a depressing image from

85
00:05:16,162 --> 00:05:21,422
Bordeaux. A firefighter injured in a barricade fire with people nearby filming. Somebody who

86
00:05:21,422 --> 00:05:26,202
came to put out a fire ends up in the hospital. My broader concern is the identity crisis I think

87
00:05:26,202 --> 00:05:30,302
the West is going through. That image alone doesn't establish anybody's immigration status

88
00:05:30,302 --> 00:05:34,462
or explain the causes of the incident. But the question of what holds a country together

89
00:05:34,462 --> 00:05:38,702
deserves a serious answer. I believe importing people with incompatible political and religious

90
00:05:38,702 --> 00:05:44,262
norms while refusing to insist on assimilation is a recipe for disaster. A country needs people

91
00:05:44,262 --> 00:05:59,723
who want to be part of it and respect its laws and customs To my mind the open borders worldview promoted by Soros and his allies treats the loss of national identity as something to welcome I think that disastrous I don see how France retains its identity if expecting people to participate in it

92
00:05:59,723 --> 00:06:04,003
is treated as unreasonable. To close out the stories, Ben Affleck gave us something considerably

93
00:06:04,003 --> 00:06:07,783
more encouraging. There's a clip in today's newsletter where he explains the work behind

94
00:06:07,783 --> 00:06:11,783
his AI filmmaking company, Interpositive. He's talking about unfreezing weights,

95
00:06:11,783 --> 00:06:14,103
training open video models on cinematic footage

96
00:06:14,103 --> 00:06:16,643
and letting filmmakers train further on their own productions.

97
00:06:17,543 --> 00:06:20,203
Netflix's acquisition announcement describes footage the team shot

98
00:06:20,203 --> 00:06:21,423
on a controlled soundstage.

99
00:06:22,063 --> 00:06:24,103
They were working on problems filmmakers actually have.

100
00:06:24,543 --> 00:06:26,403
Missing shots, backgrounds, lighting.

101
00:06:27,103 --> 00:06:29,483
What impressed me is that Affleck knows what's wrong with the footage

102
00:06:29,483 --> 00:06:31,903
and can explain how the team trained the models to improve it.

103
00:06:32,503 --> 00:06:34,503
That connects directly to what Tommy and I discussed.

104
00:06:35,243 --> 00:06:37,743
Somebody with knowledge of a field can recognize a useful result

105
00:06:37,743 --> 00:06:39,503
and tell you where the tool is falling short.

106
00:06:40,423 --> 00:06:43,123
Access to open models means they can work on those shortcomings themselves.

107
00:06:43,643 --> 00:06:46,623
They don't have to wait for a closed provider to decide their problem deserves attention.

108
00:06:47,183 --> 00:06:50,423
Before we head into the weekend, one practical thing from Freedom Tech Corner.

109
00:06:51,283 --> 00:06:54,123
Compare your signal safety numbers with your family or a Bitcoin cosigner.

110
00:06:54,823 --> 00:06:58,263
Open the conversation, tap the person's name, and select View Safety Number.

111
00:06:58,823 --> 00:07:02,043
When you're together in person, compare the numbers or scan the QR code,

112
00:07:02,343 --> 00:07:03,643
then mark the contact verified.

113
00:07:04,363 --> 00:07:05,443
The in-person part matters.

114
00:07:06,103 --> 00:07:08,103
The app doesn't independently establish somebody's identity.

115
00:07:08,683 --> 00:07:11,523
You're connecting that encrypted conversation to a person you actually know.

116
00:07:12,183 --> 00:07:13,363
If the number changes unexpectedly,

117
00:07:13,903 --> 00:07:16,783
check through an already established channel before acting on a sensitive request.

118
00:07:17,503 --> 00:07:19,843
A change can have an ordinary explanation, like a new phone.

119
00:07:20,423 --> 00:07:23,063
Don't ignore it, and don't assume it automatically means an attack.

120
00:07:23,743 --> 00:07:26,223
Verification also won't tell you whether somebody's device is compromised

121
00:07:26,223 --> 00:07:27,503
or they're being coerced.

122
00:07:28,063 --> 00:07:29,623
You still need to be willing to stop and check.

123
00:07:30,323 --> 00:07:32,483
That's your Bitcoin Brief for Friday, October 2nd.

124
00:07:32,483 --> 00:07:35,023
You'll find all the links, sources, and the full written analysis

125
00:07:35,023 --> 00:07:36,463
in today's edition of The Commoner.

126
00:07:36,463 --> 00:07:43,563
subscribe free at tftc.io slash bitcoin dash brief have a great weekend freaks

127
00:07:43,563 --> 00:07:46,543
stay humble stack sats i'll see you monday