Back Original

LLMs: WOWs & WTFs

I want to offer the framing I try to take on interactions with products like ChatGPT and Claude that help me make sense of moments of surprise, both positive and negative, or WOWs and WTFs.

The first frame is that these products are just natural language search, in contrast to key-word search, that you can not always count on. If they’ve seen enough content on a thing, and decent content on a thing, then they will give you more content on a thing. If they have not, they’ll give you anything that’s related, even if it makes no sense at all! They can not tell when they’re wrong. That’s a WTF moment. Is it wrong or just best effort retrieval? How about a WOW moment? Those “net new” discoveries it made that stumbled a field for decades but turns out hidden in plain sight1, for example, it just spliced content for parts A-B from source S, with content for parts D-E from source T, with yet more content for parts F-J from source U, to get you from part A in your request to part J in the response. Is that a discovery or just fancy retrieval?

The second frame is that these products are translators. We’ve had DSLs, compilers, and source to source translators of various quality for eons. We’ve had natural language translation for a decade at least. These products will translate from natural language to code, or from code to natural language, or indeed from one programming language to another. They can do this more arbitrarily than anything we’ve seen before (at least to my knowledge). “Product” people often point to the idea that they use these products to generate (the code for) an artifact from a big long (perhaps incomplete and buggy) specification. Is that engineering or just translation? I wrote the opening sentence in my first post hypothetically, but it seems these products really have become the primary way some people in some professions interact with computers.

Natural language is just one angle at any phenomenon2. Others include formulae, diagrams, tables, code, etc. Have these products (finally) made natural language (a bit) more tractable for computers or just made the content on the WWW more easily but not reliably accessible3?

These aren’t informed by data. They’re simply framing I use to help me understand my own experience, others’s experience, and the chasm between the two. I hope they help you too.