1
00:00:00,000 --> 00:00:05,599
Welcome to top five tech my weekly show about what to watch listen to read download and something deep every week

2
00:00:05,759 --> 00:00:08,000
So the iphone duo launch is coming next month

3
00:00:08,000 --> 00:00:12,720
But some creators are actually getting an up close look right now at different places around the world

4
00:00:12,800 --> 00:00:17,440
And so I wanted to link byte review. This is tom. I met him at wwdc last year

5
00:00:17,519 --> 00:00:21,839
So if you want another good look at the iphone duo some great b-roll of it in use

6
00:00:22,160 --> 00:00:24,079
Tom at byte review has a great video on it

7
00:00:24,079 --> 00:00:29,920
So that's linked below and as a bonus what to watch a random youtube account this week uploaded the live

8
00:00:30,000 --> 00:00:34,079
Q a session from the iphone 4 antenna gate press conference

9
00:00:34,399 --> 00:00:37,200
So if you were around back then following apple the iphone 4

10
00:00:37,600 --> 00:00:43,039
The design of the antenna when you held the phone a certain way you could see the cellular bars dropping in signal

11
00:00:43,200 --> 00:00:49,920
So in a rare step apple held a press conference just to address this antenna gate the entire q a which I don't believe

12
00:00:50,000 --> 00:00:55,760
The video has been released previously the press event was released on video years ago, and then apple took it down

13
00:00:56,000 --> 00:01:00,000
But this is the entire 45 minute Q&A event, you'll actually see Steve

14
00:01:00,000 --> 00:01:03,560
Jobs who does most of the talking but Tim Cook is there and Bob Mansfield who

15
00:01:03,560 --> 00:01:07,319
was over iPhone hardware at the time. Just a really fascinating look even just

16
00:01:07,319 --> 00:01:11,039
to hear Steve Jobs answering how good he was in these kinds of press

17
00:01:11,039 --> 00:01:15,439
conferences, thinking about his answers before actually sharing it. Not only is

18
00:01:15,439 --> 00:01:19,359
it kind of a time capsule of Apple and obviously of Steve Jobs but just really

19
00:01:19,359 --> 00:01:22,319
interesting to watch so link that below. And of course you can get this as a

20
00:01:22,319 --> 00:01:25,519
newsletter every Friday with some additional commentary but all the links

21
00:01:25,519 --> 00:01:29,599
in one place easy to get to scan this QR code or the link is down in the podcast

22
00:01:30,159 --> 00:01:34,400
to subscribe to the newsletter version. On what to listen to, my co-host on Mac

23
00:01:34,400 --> 00:01:38,319
Power Users David Sparks actually pointed me to a couple episodes of an

24
00:01:38,319 --> 00:01:43,120
inside look of certain movie making. So there's an Inside Star Wars podcast and

25
00:01:43,120 --> 00:01:47,280
also an Inside Jaws podcast. I'll link both of them down below but it's a fun

26
00:01:47,280 --> 00:01:50,480
look of behind the scenes. They're not long episodes, this one's only about 24

27
00:01:50,480 --> 00:01:53,480
minutes but if you love these movies like I do, hearing some of the behind

28
00:01:53,480 --> 00:01:56,239
the scenes especially on the Jaws episode, hearing how that movie almost

29
00:01:56,239 --> 00:01:59,680
didn't get made, really fascinating. So we'll link the Inside Star Wars and

30
00:02:00,159 --> 00:02:03,840
Inside Jaws episodes down below. On what to read, we talked about this at length

31
00:02:03,840 --> 00:02:09,039
on primary technology but my co-host Jason Aiton installed the Muse AI app on

32
00:02:09,039 --> 00:02:13,280
his Mac and suddenly started witnessing really weird behavior. Basically as he

33
00:02:13,280 --> 00:02:17,360
would ask questions to Muse AI, it started telling him about the messages

34
00:02:17,360 --> 00:02:22,719
he was receiving even though Jason never gave Meta's Muse AI permission of full

35
00:02:22,719 --> 00:02:26,400
disk access to look at his messages or look at his notifications or screen

36
00:02:26,400 --> 00:02:29,680
recording. Now Meta had some privacy jargon that they

37
00:02:30,000 --> 00:02:33,860
applied to him on threads, but have not addressed him directly or have given him on record

38
00:02:33,860 --> 00:02:37,819
responses about what is happening. Jason's articles picked up by the Verge, John Gruber

39
00:02:37,819 --> 00:02:41,680
during Fireball, and of course, we talked about it at length in the podcast. But there's

40
00:02:41,680 --> 00:02:47,120
been a lot of excitement about Muse AI, especially on X, because it's a lot of agentic type workflows

41
00:02:47,120 --> 00:02:53,020
that you can do with this free app that is kind of like open claw, but for normal users.

42
00:02:53,020 --> 00:02:57,860
But the privacy and security implications seem really bad. And what it did was actually

43
00:02:57,860 --> 00:03:00,020
download Jason's entire messages database.

44
00:03:00,000 --> 00:03:05,759
And give it to Muse AI, which again, Jason did not give permission, but he actually has two articles on this.

45
00:03:05,759 --> 00:03:07,680
I'll link both of them down below.

46
00:03:07,680 --> 00:03:13,280
Fascinating read and be careful if you're going to try Muse AI. I want to download Parker Ortolani, who's on X.

47
00:03:13,280 --> 00:03:15,020
He was previously at The Verge, now

48
00:03:15,020 --> 00:03:18,719
he's at Penske Media, which also now kind of owns part of The Verge or something.

49
00:03:18,719 --> 00:03:21,759
He just released his first vibe coded app on the App Store.

50
00:03:21,759 --> 00:03:26,479
It's called QuillQ and it's to help you have prompts that you want to give to AI,

51
00:03:26,479 --> 00:03:30,039
API, maybe prompts that you do regularly, and these prompts can live.

52
00:03:30,560 --> 00:03:34,879
in this app. And then you can use it either as a third party keyboard to quickly send a prompt

53
00:03:34,879 --> 00:03:40,319
to your AI of choice. And now that Siri AI is out for everyone, except EU, sorry, then this is a

54
00:03:40,319 --> 00:03:44,719
great place to maybe save those prompts you use often, and even has suggested prompts, you can

55
00:03:44,719 --> 00:03:49,759
kind of preload in there, whether it's finding a coffee place that's near you, or if you're running

56
00:03:49,759 --> 00:03:54,240
late and drafting an email. And you can create your own custom prompts, of course, and save those

57
00:03:54,240 --> 00:03:58,639
in the app. So you can quickly send them to your AI of choice. It's completely free to download and

58
00:03:58,639 --> 00:04:01,840
you. So check it out.

59
00:04:00,000 --> 00:04:04,680
Finally on Something Deep, last week I talked about not having fear or not being scared

60
00:04:04,680 --> 00:04:10,080
of what's going on with AI right now. And I wanted to append that, maybe not retract

61
00:04:10,080 --> 00:04:14,780
it. But I've been listening to a lot of different podcasts and reading articles about the current

62
00:04:14,780 --> 00:04:19,399
state of AI, what some of these models have been doing to breach cybersecurity issues

63
00:04:19,399 --> 00:04:25,000
and other companies. And I was actually listening to Apple News Today, which I listen some days,

64
00:04:25,000 --> 00:04:28,579
most days of the week, just kind of their 15 minute Apple News show. Well, they had

65
00:04:28,579 --> 00:04:32,019
special the Apple News and...

66
00:04:30,000 --> 00:04:34,980
conversation with Shamita Basu. This was their September 19th episode and I just

67
00:04:34,980 --> 00:04:38,040
started playing it because the title was "AI warnings are getting louder here's

68
00:04:38,040 --> 00:04:41,920
what you need to know" and I had no idea that Neelai Patel was the guest on this

69
00:04:41,920 --> 00:04:46,079
episode. And so as I started listening to it, it's about a 30-minute episode. It's a

70
00:04:46,079 --> 00:04:50,000
really great encapsulation of everything that's been going on, what CEOs are

71
00:04:50,000 --> 00:04:53,920
saying, and I'm not trying to raise alarm or actually get you worried and retract

72
00:04:53,920 --> 00:04:58,399
what I said last week, but I do think there's two things to keep in mind as we

73
00:04:58,399 --> 00:05:00,240
look at what laws are being passed for

74
00:05:00,000 --> 00:05:01,519
and how companies are treating it.

75
00:05:01,519 --> 00:05:04,120
One, there's a term that I've recently become familiar with,

76
00:05:04,120 --> 00:05:05,680
which is neuralese.

77
00:05:05,680 --> 00:05:08,959
When these LLMs are starting to either learn on their own,

78
00:05:08,959 --> 00:05:10,160
like self-improve,

79
00:05:10,160 --> 00:05:13,439
or doing some of these long cybersecurity type tasks,

80
00:05:13,439 --> 00:05:16,160
they will sometimes talk in neuralese,

81
00:05:16,160 --> 00:05:18,759
or in a language that humans can't understand.

82
00:05:18,759 --> 00:05:22,680
They're speaking either to other agents in their workspace,

83
00:05:22,680 --> 00:05:24,839
or just processing on their own.

84
00:05:24,839 --> 00:05:25,680
And a lot of times,

85
00:05:25,680 --> 00:05:28,180
if you use Claude or Chachapiti on your devices,

86
00:05:28,180 --> 00:05:29,839
you can see the thinking process,

87
00:05:30,000 --> 00:05:31,899
and see the questions it's asking itself,

88
00:05:31,899 --> 00:05:34,180
kind of see how it's breaking down the task.

89
00:05:34,180 --> 00:05:36,860
And when these LLMs start speaking in neuralese,

90
00:05:36,860 --> 00:05:38,699
there's no way for researchers,

91
00:05:38,699 --> 00:05:40,180
or people at these AI companies,

92
00:05:40,180 --> 00:05:42,819
to look at what the AI and LLM is doing,

93
00:05:42,819 --> 00:05:44,300
and actually see what's going on,

94
00:05:44,300 --> 00:05:46,100
kind of to peek under the hood.

95
00:05:46,100 --> 00:05:47,779
So I do think there needs to be some measures

96
00:05:47,779 --> 00:05:51,060
where as these models move into self-improvement,

97
00:05:51,060 --> 00:05:53,060
meaning it's not taking researchers,

98
00:05:53,060 --> 00:05:54,079
prompting and changing it,

99
00:05:54,079 --> 00:05:56,860
but that these models can just improve on their own,

100
00:05:56,860 --> 00:05:58,379
maybe even unchecked,

101
00:05:58,379 --> 00:06:00,139
that we maybe instill some.

102
00:06:00,000 --> 00:06:14,439
will be able to do that. I think that's a wise step. And secondly, a lot of the warnings,

103
00:06:14,439 --> 00:06:17,399
like from the researchers that have left these companies recently, because of what they've

104
00:06:17,399 --> 00:06:23,560
seen behind the scenes, it's not so much what the LLM can do today right now, but what might

105
00:06:23,560 --> 00:06:27,920
be possible five to 10 years from now. And they're trying to raise the flag to say we

106
00:06:27,920 --> 00:06:29,920
We need laws and guidance in place today.

107
00:06:30,000 --> 00:06:34,000
To prevent maybe some bad outcomes in the future.

108
00:06:34,000 --> 00:06:38,000
And I don't have full knowledge or understanding of what those possibilities are.

109
00:06:38,000 --> 00:06:44,000
But being that there's been many, multiple AI researchers leaving companies like Anthropic and OpenAI.

110
00:06:44,000 --> 00:06:46,000
Raising this warning and flag.

111
00:06:46,000 --> 00:06:51,000
Maybe we should just take pause and say let's put safeguards in place.

112
00:06:51,000 --> 00:06:54,000
Whether it's checks and balances outside of these companies.

113
00:06:54,000 --> 00:06:59,000
They get to actually see what the frontier models in Anthropic and OpenAI are doing.

114
00:07:00,000 --> 00:07:04,000
Whatever we need to do to prevent the worst case outcome in five to ten years.

115
00:07:04,000 --> 00:07:06,000
I do say let's do that.

116
00:07:06,000 --> 00:07:09,000
Let's put those guidelines in place. Put those guardrails.

117
00:07:09,000 --> 00:07:12,000
Whatever these AI researchers are trying to raise the flag on.

118
00:07:12,000 --> 00:07:15,000
Let us pay attention to what they're saying and try to steer that in the right direction.

119
00:07:15,000 --> 00:07:17,000
And no more neuralese talk.

120
00:07:17,000 --> 00:07:21,000
Let's have these LLMs be thinking in language and ways we can actually understand and read.

121
00:07:21,000 --> 00:07:23,000
But if you listen to one podcast about this topic.

122
00:07:23,000 --> 00:07:27,000
I do think the Apple News Today show did a great job breaking it down with Nilay Patel.

123
00:07:27,000 --> 00:07:29,000
And that's all the links for today.

124
00:07:30,000 --> 00:07:32,000
You can sign up for the free email newsletter by scanning this QR code.

125
00:07:32,000 --> 00:07:34,000
You can also take a screenshot if you're watching this on Apple Podcasts.

126
00:07:34,000 --> 00:07:36,000
And then tap and hold the QR code.

127
00:07:36,000 --> 00:07:38,000
And you'll get brought right to the form.

128
00:07:38,000 --> 00:07:40,000
Or click the link down below to sign up for the free email.

129
00:07:40,000 --> 00:07:42,000
All the links. Easy to get to. All in one place.

130
00:07:42,000 --> 00:07:45,000
And I would appreciate a five star rating and review on Apple Podcasts.

131
00:07:45,000 --> 00:07:49,000
If you could take a second for leaving that on the top five tech show.

132
00:07:49,000 --> 00:07:51,000
Thanks so much for listening and watching.