Stochastic Parrots Do Not Understand Þt Hexapodia Is, in Fact, þe Key Insight

How Chat-GPT boosters are deluding þemselves in thinking þt it is much more þan it is—but þt does not keep it from being, potentially, very useful indeed. A stochastic parrot is a useful assistant...

How Chat-GPT boosters are deluding þemselves in thinking þt it is much more þan it is—but þt does not keep it from being, potentially, very useful indeed. A stochastic parrot is a useful assistant because so much of our use of language is in fact (also[?]) stochastic parrotage…

---

As I have been periodically saying for pretty much the past year, my big first-order takeaway from Chat-GPT in particular and recent work in GPT-LLM-ML in general is not that these models are in any sense a road to AGI, but rather that their usefulness stems from how much of our use of language is actually not AGI either—that we are, in our daily activities, largely stochastic parrots too.

And so a stochastic parrot can easily be a very helpful virtual assistant. But—only as a stochastic parrot.

And, this morning, something comes across my desk that makes me want to say this again:

The smart Ben Levinstein writes:

Ben Levinstein: How to Think about Large Language Models: ’If you hang out on social media at all, you sometimes see (non-ML) professors worried about the future of teaching say things like, “Well, ChatGPT can write semi-coherent prose, but it can’t do [x],” where [x] means something like write creatively, solve mildly complex math problems, generate interesting new ideas, pass some college level exam, or whatever. These professors are right, and ChatGPT does not have those capabilities. But it’s generally not clear that we need any kind of conceptual innovation to get the next chatbot to get to that level. We may just add some more layers and parameters and training and just generally throw a bit more compute at our model next time, and voilà…

What can Chat-GPT actually do?

Here is my latest attempt to find out:

---

Some backstory: Noah Smith proposed and I agreed that we take a particular phrase—“hexapodia is the key insight”—and make it the title of our podcast. Its source is Vernor Vinge’s A Fire Upon the Deep  <https://archive.org/details/fireupondeep0000ving>. And the meaning we ascribe to it in the context of our podcast is, as we have now said 52 times, once in every podcast episode: That there is often some key nugget of fact that—if you understand it correctly and place it in its proper context—will transform your view of the situation, and allow you to grok it correctly.

Google comes up with only 53 hits total for the search ‘“Brad DeLong” AND “hexapodia”’, all of which are links to or references to Noah’s and my podcast <https://braddelong.substack.com/s/hexapodia-is-the-key-insight-by-noah>, in every single one of which we provide this explanation of why we use the phrase: “hexapodia is the key insight”.

An internet search-enabled version of Chat-GPT—which is what Bing Chat is—should be able to pick this up if asked any question involving “Brad DeLong” AND “hexapodia”, clothe it in appropriate prose, and return it as an answer no?

But it does not.

This is what happens when we ask Chat-GPT4 in its Bing Chat guise the question: “Does Brad DeLong believe that hexapodia is the key insight?”

Searching for: Brad DeLong hexapodia

Generating answers for you…

Hello, this is Bing. I can help you find some information about Brad DeLong and hexapodia. 😊…

…Hexapodia is a term coined by Brad DeLong[1], a professor of economics at UC Berkeley, and Noah Smith, a columnist for Bloomberg Opinion[2], for their podcast series. The name is inspired by a science fiction story by Vernor Vinge, where an alien race of six-legged creatures[3] has a different way of thinking[4] than humans. The podcast explores various topics in economics, finance, political economy, and history, with an emphasis on the key insights that shape our understanding of the world.

Brad DeLong believes that hexapodia is the key insight in the sense that it represents the importance of being open-minded and curious[5] about different perspectives and paradigms. He often uses the phrase “hexapodia is the key insight” as a humorous way of introducing a new or surprising idea that challenges conventional wisdom or common sense[6]. For example, in one episode, he says that “hexapodia is the key insight” to explain why China’s industrial policy is successful and unique[7]. In another episode, he says that “hexapodia is the key insight” to argue that the Biden administration’s economic policies are not inflationary[8] and are based on sound macroeconomic principles.

Therefore, Brad DeLong does not literally believe that hexapodia is the key insight, but rather uses it as a rhetorical device to invite listeners to think outside the box and question their assumptions[9]. He and Noah Smith also use it as a catchy title for their podcast series, which aims to provide interesting and informative discussions on various topics related to economics and society.[10]

[1] I did not coin the term hexapodia.
[2] Noah Smith is not a columnist for Bloomberg any more.
[3] The alien creatures do not have six legs, but rather ride on scooters with six wheels.
[4] Actually, the alien creatures think—when they have had their short-term memory-prostheses activated—in very human ways.
[5] No: it means not that you should be more open-minded, but that you should look very carefully at what you already see.
[6] No: we use it once at the beginning of each podcast, and once at the end. We do not use it to introduce “new or surprising ideas”.
[7] I did not do this.
[8] I did not do this.
[9] I do not do this.
[10] This is correct—except for the “also”, which is wrong. This is our sole use of it. This is not an “also” use.

---

Now from my perspective, this simply will not do as an answer.

Chat-GPT4, in its Bing Chat guise,is supposed to stay grounded in reality—or, at least, grounded in the internet—via web searches. The only places it could find the terms “Brad DeLong” AND “hexapodia” should lead it to our explicit statement of what we mean by “hexapodia is the key insight”.

It does not.

What does it do instead?

1. Note what it does not do. It does not answer the question.

2. Instead of answering the question, it says “I can help you find some information about Brad DeLong and hexapodia 😊”. Now this is something that people sometimes do when they are dodging a question, and thus it is in the training set as a human reaction.

3. It then produces stochastic-parrotage word-salad, with seven incorrect and only two correct sentences, which are: “The podcast explores various topics in economics, finance, political economy, and history, with an emphasis on the key insights that shape our understanding of the world.” and “He and Noah Smith also use it as a catchy title for their podcast series, which aims to provide interesting and informative discussions on various topics related to economics and society.” (But, here in the second sentence, the “also” is incorrect.)

Now do not get me wrong: I think these GPT models will be very useful, because a huge proportion of what we write is, in fact, semi-ritualistic stochastic parrotage. In writing first drafts of such, and in reading such writing in order to extract the Shannon-information gist, I am expecting to use GPT’s a lot over the next decade to make my life easier.

But that is all that it is.

The answer that it should have given is:

Hello, this is Bing. I can help you find some information about Brad DeLong and hexapodia. 😊 Brad DeLong and Noah Smith use “hexapodia is the key insight” as a catchy title for their podcast series, which aims to provide interesting and informative discussions on various topics related to economics and society. They do not literally believe that “hexapodia is the key insight”. But in the context in which the phrase is found in Vernor Vinge’s novel A Fire Upon the Deep, “hexapodia” is a key nugget of fact that—had the character understood it correctly and placed it in its proper context—would have transformed their view of the situation, and allowed you to grok it correctly. And they title their podcast “Hexapodia is the key insight” out of their belief that this is generally true.

I do not see how you get there from the word-salad stochastic-parrotage that was its initial answer without both a theory of the world and a theory of mind. And I do not see how the current training régimen is going to get there. It seems to me to requires much more and very different from merely “add[ing] some more layers and parameters and training and just generally throw a bit more compute at our model next time…”

---

Background on the Term “Hexapodia” in Vernor Vinge’s A Fire Upon the Deep:

About one-sixth of the way through Vernor Vinge’s space opera A Fire Upon the Deep, one of the protagonists—a human named Ravna Bergsndot—is scanning… USENET… I guess these days one should call it Discord… for messages about the Threat of the “Blight”, which is a hyperintelligent civilization-gobbling mind virus that the civilizations of the galaxy are, well, threatened by. This Blight has already gobbled a number of planets and a major node, Relay, of the galaxy’s interstellar communications network. 

An important clue—it says, explicitly, that it is a “key insight”—passes her by. She fails to pick up on it because of (a) the weirdness (relative to her) of the weltenschauung of the speaker, and (b) consequent difficulties in translation. 

What is this key insight she misses? 

“Hexapodia”, of course:

Ravna Bergsndot starts to read the post:

Twirlip of the Mists [Perhaps an organization of cloud fliers in a single jovian system. Very sparse priors.]
Subject: Blighter Video thread Key phrases: Hexapodia as the key insight Distribution: Threat of the Blight
Date: 8.68 days since Fall of Relay
Start of text of message: I haven’t had a chance to see the famous video from Straumli Realm, except as an evocation. (My only gateway onto the Net is very expensive.) Is it true that humans have six legs? I wasn’t sure from the evocation. If these humans have three pairs of legs, then I think there is an easy explanation for —…

She does not click to expand and see the full message.

She, rather, dismisses it, as either a translation error or as coming from some highly bewildered and essentially clueless creature.

Twirlip of the Mists shows up later on in the book as well:

Twirlip of the Mists [Perhaps an organization of cloud fliers in a single jovian system. Very sparse priors before this thread began. Appears to be seriously out of touch. Program recommendation: delete this poster from presentation.]
Subject: The Blight’s goal at the Bottom
Distribution: Threat of the Blight Great Secrets of Creation
Date: 4.54 days since Fall of Sjandra Kei
Key phrases: Zone Instability and the Blight, Hexapodia as the key insight
Text of message: Apologies first if I am repeating obvious conclusions. My only gateway onto the Net is very expensive, and I miss many important postings. I think that anyone following both Great Secrets of Creation and Threat of the Blight would see an important pattern. Since the events reported by Harmonious Repose information service, most agree that something important to the Perversion exists at the Bottom of the Beyond in region […]. I see a possible connection here with the Great Secrets. During the last two hundred and twenty days, there have been increasing reports of zone interface instability in the region below Harmonious Repose. As the Blight threat has grown and its attacks against advanced races and other Powers continued, this instability has increased. Could there not be some connection? I urge all to consult their information on the Great Secrets (or the nearest archive maintained by that group). Events such as this prove once again that the universe is all ronzelle between…

And:

Twirlip of the Mists [Who knows what this is, though probably not a propaganda voice. Very sparse priors.]
Subject: The cause of the recent Great Surge Distribution: Threat of the Blight Great Secrets of Creation Zonometric Interest Group
Date: 66.47 days since Fall of Sjandra Kei
Key phrases: Zone Instability and the Blight, Hexapodia as the key insight
Text of message: Apologies if I am repeating obvious conclusions. My only gateway onto the Net is very expensive, and I miss many important postings. The Great Surge now in progress appears by all accounts to be an event of cosmic scope and rarity. Furthermore, the other posters put its epicenter less than 6,000 light-years from recent warfare related to the Blight. Can this be mere coincidence? As has long been theorized [citations from various sources, three known to Ølvira ; the theories cited are of long standing and nondisprovable] the Zones themselves may be an artifact, perhaps created by something beyond Transcendance for the protection of lesser forms, or [hypothetical] sentient gas clouds in galactic cores. Now for the first time in Net history we have a Transcendent form, the Blight, that can effectively dominate the Beyond. Many on the Net [cites Hanse and Sandor at the Zoo] believe that it is searching for an artifact near the Bottom. Is it no wonder that this could upset the Natural Balance and provoke the recent Event? Please write to me and tell me what you think. I don’t get much mail…

In his note on the first of Twirlip’s USENET posts, author Vernor Vinge writes, apropos of six legs:

Heh, heh: … or three pairs of wheels? I wonder what would have happened if Ravna had just read a little further. In some weird way, Twirlip knows the Secret of the Riders. I wonder how many people will catch this. It’s really not up to the level of a legitimate clue (I didn’t notice it until after I wrote it) — but if it were, Ravna would have instantly caught on to it. This is a special case of something you might use elsewhere: Even though the Known Net has enormous connectivity, the interests of the participants and the prejudices of the newsfilter software would tend to create virtual partitions. There could be large segments that, sometimes unknowingly, are ignoring each other. Most of the time this would just improve efficiency; in some cases great insights would be lost. (Hence, I bet some people or their automation would expend lots of effort dredging the unintelligible. Even that would not eliminate the problem.)…

In his note on the second:

You might remark that Twirlip is getting in because of “open-mindedness” options in the newsreader automation.,,

And in his note on the third:

Twirlip may know all the big secrets, but it’s just too strange for credible translations — hence its permanent crank status throughout this story…