There is a duality at the heart of large language models. On the one hand, they are essentially a backwards-looking invention, a “cultural technology”, in the words of Alison Gopnik—algorithms which index and remix a large slice of human culture (though one that is typically heavily biased towards the recent past). On the other hand, they can often seem to be producing something entirely new, and can thereby leave many people with the impression of having a personality or even “sentience” (whatever that means exactly); in the most extreme cases, some people have apparently convinced themselves that such models are a step on the path towards some sort of successor species to humanity, a new regime of algorithmic children that will survive our own human catastrophes. Complicating matters here is the fact that the emergence of and widespread attention to these systems largely overlapped with the Covid-19 pandemic, a time in which we have all had additional reason to reflect on life, death, loss, and creation.

Tags:  machine-learning ,  large-language-models ,  archives ,  books ,  fiction ,  memory ,  mortality ,  digital-preservation

Tags:  large-language-models ,  chatgpt ,  reproducibility ,  evaluation ,  open-science

Tags:  visualization ,  altair ,  bokeh

It was recently reported by inews.co.uk that Twitter is going to start charging academic researchers and institution $42,000/month if they want to maintain their current level of expansive access to data, and – more significantly – require that they delete all Twitter data from their archives if they do not. I’d heard rumors that this might be happening a few weeks ago, but the iNews article is the first independent reporting that I’ve seen about it.

Tags:  twitter ,  archives ,  sociotechnical-systems ,  reproducibility ,  change ,  instability

ChatGPT has understandably garnered a huge amount of attention from all corners of academia, from philosophy to economics. One of the more quixotic examples I’ve encountered recently is Robert W. McGee, and his many papers on this topic, such as “Is Chat Gpt Biased Against Conservatives? An Empirical Study”.

A professor of accounting at Fayetteville University, McGee’s biography reads as something like a Marvel Cinematic Universe version of a nerdy academic supervillain. In addition to having published 59 non-fiction books, McGee apparently holds 23 academic degrees, including 13 doctorates, as well as being a world champion in various martial arts, such as Taekwondo and Tai Chi.

Tags:  large-language-models ,  chatgpt ,  sociotechnical-systems ,  measurement ,  evaluation

Last week, a number of US outlets reported on data leaks at Samsung, in which three employees (in separate incidents) apparently entered confidential company information into ChatGPT. According to reports, in one case, an employee tried using ChatGPT to help debug code, another to optimize code, and a third to have it produce a summary of meeting notes.

Tags:  machine-learning ,  large-language-models ,  chatgpt ,  data ,  privacy ,  security

I’ve written about this before, but it’s worth remembering that almost nothing in sociotechnical systems is guaranteed to remain stable for very long. We’ve recently had two great examples of this, with the first being the changes to Twitter, and the second being ChatGPT (and, by extension, the new Bing).

In the first case, a platform which had long seemed relatively static, (especially compared to all the rest), rather suddenly changed hands, which led to major changes in what it delivered. On some level, many of the technical changes to the actual functionality of the site were relatively minor. Much bigger, however, was the impact of many people abandoning the platform for alternatives like Mastodon. Although it seems to me like people have gradually been filtering back to Twitter, most people seem to have the sense that the experience has changed. Obviously more dramatic infrastructural changes, like prioritizing tweets from paying users, could produce even more dramatic shifts. Regardless, it’s a good reminder that what we think of as “Twitter” is the product of a combination of people and software, either or both of which can shift dramatically in a short period of time.

Tags:  large-language-models ,  twitter ,  sociotechnical-systems ,  chatgpt ,  instability ,  archives

I expect that almost anyone reading this will have heard of ChatGPT by now. Released about a month ago, ChatGPT is a system developed by OpenAI which provides text responses to text input. Although details are scarce, under the hood ChatGPT is basically a large language model, trained with some additional tricks (see Yoav Goldberg’s write up for a good summary). In other words, it is a model which maps from the text input (treated as a sequence of tokens), to a distribution over possible next tokens, and generates text by making repeated calls to this function, and sampling tokens from the predicted distributions.

Tags:  machine-learning ,  large-language-models ,  chatgpt ,  open-science ,  evaluation

Tags:  artificial-intelligence ,  machine-learning ,  algorithmic-decision-making ,  governance ,  software ,  large-language-models ,  sociotechnical-systems ,  power

Sep 23, 2022

Hacking LLM bots

For anyone who missed it, a Twitter account named @mkualquiera recently deployed what seems like a kind of adversarial attack in the wild on a large language model (LLM)-based Twitter bot. I’ll link to the key post below, but it’s worth providing a bit of context, as it wasn’t immediately clear to me what was going on when I first saw the tweet.

Tags:  machine-learning ,  large-language-models ,  twitter ,  bots ,  sociotechnical-systems