WEBVTT - Monologue: No, Something Big Isn't Coming

0:00:02.800 --> 0:00:06.800
<v Speaker 1>Zo Media, Hello, and welcome to Better Offline. This is

0:00:06.840 --> 0:00:09.200
<v Speaker 1>your monologue for the week, and I'm your host ed zeitron.

0:00:17.280 --> 0:00:19.959
<v Speaker 1>So you've probably seen, or heard about, or even read

0:00:20.000 --> 0:00:24.320
<v Speaker 1>the abominable, nonsensical hog slob essay called something Big is Coming,

0:00:24.520 --> 0:00:27.320
<v Speaker 1>perhaps seeing people that should know better, like Mehdi has

0:00:27.320 --> 0:00:31.000
<v Speaker 1>Sons who called it the most important thing you read today.

0:00:31.240 --> 0:00:34.640
<v Speaker 1>Despite it being both AI generated and complete fucking nonsense,

0:00:35.120 --> 0:00:37.600
<v Speaker 1>this essay has done the rounds because of its historyonic

0:00:37.680 --> 0:00:40.600
<v Speaker 1>tone and deliberate leaning upon the anxieties of those scared

0:00:40.600 --> 0:00:43.280
<v Speaker 1>of AI and the mythology of AI boosters, and across

0:00:43.320 --> 0:00:46.720
<v Speaker 1>forty seven hundred b shit ass words, says very little

0:00:46.760 --> 0:00:50.160
<v Speaker 1>new or interesting, along with multiple egregious lies. I will

0:00:50.159 --> 0:00:53.519
<v Speaker 1>be attaching an annotated version of this essay to this monologue,

0:00:53.560 --> 0:00:55.200
<v Speaker 1>but I want to give you a broad explanation of

0:00:55.240 --> 0:00:57.520
<v Speaker 1>what it really says and why it's full of shit.

0:00:57.960 --> 0:01:00.600
<v Speaker 1>But let's start with who wrote m that. Schumer is

0:01:00.600 --> 0:01:03.000
<v Speaker 1>a guy who makes an AI startup called other Side Ai,

0:01:03.080 --> 0:01:05.440
<v Speaker 1>which does this thing called hyperwrite, which is an AI

0:01:05.600 --> 0:01:09.360
<v Speaker 1>writing assistant with web search and citations, which translates to

0:01:09.440 --> 0:01:12.399
<v Speaker 1>mean it's a rapper for various large language models that

0:01:12.440 --> 0:01:15.280
<v Speaker 1>he's prompting appropriately. When I asked it to write a

0:01:15.319 --> 0:01:18.080
<v Speaker 1>critique of his piece, it's said that its weaknesses lied

0:01:18.160 --> 0:01:21.600
<v Speaker 1>in its lack of specifics, and that its alarmist framing

0:01:21.640 --> 0:01:23.920
<v Speaker 1>grabs attention, and it's heavy on the hype and light

0:01:23.959 --> 0:01:26.880
<v Speaker 1>on actual insight. And you know what, Matt, you fucking

0:01:27.160 --> 0:01:30.000
<v Speaker 1>You've got one innovation, and it's that even your shit

0:01:30.120 --> 0:01:33.399
<v Speaker 1>ast lm rapper doesn't like your bullshit. And I think

0:01:33.440 --> 0:01:37.360
<v Speaker 1>it's also being too kind. Schumer makes numerous egregious misstatements

0:01:37.400 --> 0:01:40.160
<v Speaker 1>about the current state of AI, using vague yet scary

0:01:40.200 --> 0:01:42.480
<v Speaker 1>statements like being able to walk away from the computer

0:01:42.520 --> 0:01:45.039
<v Speaker 1>for hours and find his work done again. His company

0:01:45.080 --> 0:01:47.600
<v Speaker 1>has done the same thing for years, although an unnamed

0:01:47.680 --> 0:01:50.120
<v Speaker 1>managing partner at a big law firm uses AI for

0:01:50.200 --> 0:01:53.000
<v Speaker 1>hours to do the work of a team of associates.

0:01:53.440 --> 0:01:55.720
<v Speaker 1>I choose these two statements because they're fairly typical of

0:01:55.760 --> 0:01:58.680
<v Speaker 1>his entire piece, entirely made up bullshit about how long

0:01:58.720 --> 0:02:01.120
<v Speaker 1>an AI can work for what it's capable of doing.

0:02:02.320 --> 0:02:05.200
<v Speaker 1>Schumer repeatedly talks about AI being able to write software

0:02:05.240 --> 0:02:07.559
<v Speaker 1>completely autonomously in a way that could not be further

0:02:07.600 --> 0:02:10.880
<v Speaker 1>from the truth. His most egregious one is when he

0:02:10.960 --> 0:02:14.160
<v Speaker 1>quotes open ai saying that GPT five point three codex

0:02:14.360 --> 0:02:16.880
<v Speaker 1>is the first model that was instrumental in creating itself

0:02:17.000 --> 0:02:19.600
<v Speaker 1>because it used early versions to debug its own training,

0:02:19.840 --> 0:02:23.360
<v Speaker 1>manage its own deployment, and diagnosed test results in evaluation.

0:02:23.800 --> 0:02:26.280
<v Speaker 1>These are actual quotes from open ai. He's not lying

0:02:26.320 --> 0:02:28.480
<v Speaker 1>they said them, he's just lying about what they mean.

0:02:28.960 --> 0:02:30.880
<v Speaker 1>To be clear, this is open ai saying that he

0:02:31.000 --> 0:02:33.560
<v Speaker 1>used an app it built to test and debug code,

0:02:33.560 --> 0:02:36.760
<v Speaker 1>and to test and debug the actual code and how

0:02:36.760 --> 0:02:37.320
<v Speaker 1>it went up.

0:02:38.120 --> 0:02:42.000
<v Speaker 2>It's literally, it's what the product they used. They tested

0:02:42.040 --> 0:02:44.040
<v Speaker 2>the product by testing what the product did. It's not

0:02:44.200 --> 0:02:46.799
<v Speaker 2>the same as saying that software is designing itself, which

0:02:46.840 --> 0:02:50.360
<v Speaker 2>is something that Schumer heavily intimates is happening throughout the piece,

0:02:50.440 --> 0:02:53.239
<v Speaker 2>which it is not. He also suggests multiple times that

0:02:53.320 --> 0:02:57.560
<v Speaker 2>AI is learning from itself, which is not. Barely keeping

0:02:57.600 --> 0:03:00.000
<v Speaker 2>my cool on this one because it's so fucking frustrate

0:03:00.560 --> 0:03:03.800
<v Speaker 2>seeing how many people I saw Fortune repost this, I

0:03:03.800 --> 0:03:07.880
<v Speaker 2>saw Business Insider do a post fucking the Technology Brothers. Well, well, well,

0:03:07.960 --> 0:03:10.240
<v Speaker 2>if it isn't the technology Brothers, they had this fuck

0:03:10.280 --> 0:03:14.640
<v Speaker 2>not on it genuinely infuriates me, and most of the

0:03:14.760 --> 0:03:17.239
<v Speaker 2>scary stuff in there. That all software engineers will be

0:03:17.280 --> 0:03:19.600
<v Speaker 2>replace the fifty percent of entry level white collar jobs

0:03:19.639 --> 0:03:20.040
<v Speaker 2>be gone.

0:03:19.919 --> 0:03:23.120
<v Speaker 1>In one to five years. Is Schumer directly copy pasting

0:03:23.160 --> 0:03:26.240
<v Speaker 1>stuff that CEO of Anthropic Warrio Amade said in one

0:03:26.280 --> 0:03:28.880
<v Speaker 1>of his many interviews where he makes shit up. I

0:03:28.919 --> 0:03:33.560
<v Speaker 1>am begging people to please God engage with basic critical thinking.

0:03:33.639 --> 0:03:36.440
<v Speaker 1>When reading things like this, Schumer says a lot of

0:03:36.560 --> 0:03:40.240
<v Speaker 1>vague marketing fluff. Each generation builds the next, which is smarter,

0:03:40.320 --> 0:03:43.240
<v Speaker 1>which builds the next faster, which is smarter. Still, this

0:03:43.400 --> 0:03:45.640
<v Speaker 1>is written specifically to make you believe that this is

0:03:45.680 --> 0:03:49.640
<v Speaker 1>actually happening when it is not. It isn't. The generation

0:03:49.760 --> 0:03:52.200
<v Speaker 1>is not building the next. There are still people doing this.

0:03:52.800 --> 0:03:55.280
<v Speaker 1>He doesn't cite anything, He doesn't even bother to choose

0:03:55.320 --> 0:03:58.200
<v Speaker 1>real life examples. He just says stuff like AI is

0:03:58.240 --> 0:04:01.480
<v Speaker 1>approaching or exceeding human performance in several areas of medicine

0:04:01.800 --> 0:04:05.080
<v Speaker 1>or without citation, Well, I've got a citation for you.

0:04:05.520 --> 0:04:09.320
<v Speaker 1>On September fifth, twenty twenty four, Matt Schumer released a

0:04:09.320 --> 0:04:12.520
<v Speaker 1>model called Reflection seventy B that he claimed had benchmarks

0:04:12.520 --> 0:04:14.520
<v Speaker 1>that made it and I quote, the world's top open

0:04:14.560 --> 0:04:18.560
<v Speaker 1>source AI model built with something called glave AI. Independent

0:04:18.560 --> 0:04:21.160
<v Speaker 1>researchers were unable to recreate the benchmarks, and it became

0:04:21.200 --> 0:04:23.919
<v Speaker 1>obvious that Schumer was a lying sack of shit and

0:04:23.960 --> 0:04:26.599
<v Speaker 1>the reflection seventy b was a rapper for anthropics claude

0:04:26.640 --> 0:04:29.800
<v Speaker 1>sonnet model, with one person posting that they had intentionally

0:04:29.880 --> 0:04:32.440
<v Speaker 1>made it so that it would not respond to being

0:04:32.520 --> 0:04:35.560
<v Speaker 1>called Claude or say that it was Claude. He also

0:04:35.600 --> 0:04:38.080
<v Speaker 1>didn't reveal that he was an investor in GLAVEI, the

0:04:38.080 --> 0:04:41.680
<v Speaker 1>platform that generated the synthetic data that allegedly trained this

0:04:41.800 --> 0:04:45.640
<v Speaker 1>shitass model that does not work, did not work, never updated,

0:04:45.760 --> 0:04:48.760
<v Speaker 1>never did anything it said, and everyone just memory hold it.

0:04:49.680 --> 0:04:53.119
<v Speaker 1>And what horrified me about this essay is how many

0:04:53.200 --> 0:04:56.120
<v Speaker 1>top tier media outlets and respectable journalists took it as

0:04:56.120 --> 0:04:59.240
<v Speaker 1>gospel because the statements he made vaguely lined up with

0:04:59.279 --> 0:05:02.680
<v Speaker 1>the empty high of scumbacks like Sam Altman and Dario

0:05:02.720 --> 0:05:04.839
<v Speaker 1>Amma Day. If you're a member of the media that

0:05:04.880 --> 0:05:07.840
<v Speaker 1>share to republish this, you are actively participating in an

0:05:07.839 --> 0:05:11.159
<v Speaker 1>act of disinformation and you should be fucking ashamed of yourself.

0:05:11.760 --> 0:05:14.880
<v Speaker 1>Fuck Matt Schumer and fuck anybody who takes them seriously.

0:05:15.320 --> 0:05:18.159
<v Speaker 1>Regular people got scared because of this because mainstream media

0:05:18.240 --> 0:05:21.560
<v Speaker 1>validated it. You should all be ashamed of yourselves. And

0:05:21.600 --> 0:05:24.080
<v Speaker 1>to the many haters that emailed me about this saying

0:05:24.080 --> 0:05:28.080
<v Speaker 1>this looks like bullshit, you were correct. These people are

0:05:28.200 --> 0:05:31.960
<v Speaker 1>deceiving you. They are misleading you so that they can

0:05:32.040 --> 0:05:34.760
<v Speaker 1>sell AI software that they know is not doing the

0:05:34.800 --> 0:05:37.599
<v Speaker 1>things it needs to to make any of this make sense.

0:05:38.440 --> 0:05:41.039
<v Speaker 1>And I am disgusted to see this keep happening. I

0:05:41.080 --> 0:05:43.919
<v Speaker 1>am disgusted to see people like this platformed. And I

0:05:43.960 --> 0:05:45.960
<v Speaker 1>think when all of this ends, people think that I

0:05:46.000 --> 0:05:48.760
<v Speaker 1>will forget. I will not forget any of this. I

0:05:48.800 --> 0:05:52.039
<v Speaker 1>am taking fucking detailed notes of everybody who pushes this

0:05:52.040 --> 0:05:56.320
<v Speaker 1>shit up hill. I am furious. I'm disgusted to see this.

0:05:56.600 --> 0:05:59.640
<v Speaker 1>Sixty million fucking views on this bullshit. And I know

0:05:59.680 --> 0:06:01.400
<v Speaker 1>there are some people who don't like when I get angry,

0:06:01.400 --> 0:06:04.320
<v Speaker 1>but I am furious about this. This is lies, This

0:06:04.400 --> 0:06:08.120
<v Speaker 1>is platforming lies. Anyone in the media who platform this,

0:06:08.680 --> 0:06:11.400
<v Speaker 1>you need to, I don't know, think about whether you're

0:06:11.480 --> 0:06:14.520
<v Speaker 1>capable of doing your jobs every day, because you seem

0:06:14.560 --> 0:06:19.800
<v Speaker 1>to lack the critical thinking to spot a scam artist. Yeah. Yeah,

0:06:19.839 --> 0:06:22.440
<v Speaker 1>I'm done. I'm done everyone. I'm sorry for flying off

0:06:22.480 --> 0:06:23.320
<v Speaker 1>the handle a little