> ## Content Index
> Fetch the complete content index at: https://www.bramadams.dev/llms.txt
> Use this file to discover other available public pages before exploring further.

# Notes on OpenAI Q&A Finetuning GPT-3 Vs Semantic Search - Which to Use, When, and Why
- URL: https://www.bramadams.dev/202302172335/
- Published: 2023-02-18T04:40:40.000Z
- Updated: 2023-02-18T18:19:00.000Z
- Description: A great video about finetuning vs semantic search. Finetuning teaches a model to write new patterns, not to have a theory of mind.
- Author: Bram Adams
- Tags: youtube-notes, ai

```timestamp-url
https://www.youtube.com/watch?v=9qq6HTr7Ocw

```

## Overall Thoughts

This is really great video; it was very thorough and had great analogies. I didn't know about the unfreezing of the partial model, that's a neat fact!

I've had many office hours with people coming with finetune related questions that I believe would be better off 99% of the time with semantic search and the Hypothetical Document Embeddings (HyDE) in the rarest of cases perhaps.

## Notated Transcript

```timestamp
 01:21

```

a history of transfer learning

```timestamp
 01:49

```

long live NLU, rip NLP

```timestamp
 02:30

```

fine tuning is tweaking a task

```timestamp
 03:46

```

only similarity b/w finetuning is q/a search is that they both use embeddings at some point

```timestamp
 04:34

```

fine tuning unfreezes part of a model -- does not stop confabulation (hallucination)

```timestamp
 05:30

```

unfreezing an entire model is expensive af

```timestamp
 06:00

```

models barf out patterns, they do not have a theory of mind or knowledge

bigger models are more convincing, but a largest model will never know itself (as an information store)

```timestamp
 08:20

```

finetuning is way more difficult than prompt engineering (10,000x harder)

```timestamp
 09:20

```

finetuning at scale is very hard -- how much do we share in alignment

```timestamp
 11:11

```

cost of fine tuning goes up with more data -- needs constant retraining

```timestamp
 12:20

```

instruct -> question + body of info -> is answer in here?

```timestamp
 12:58

```

finetuning teaches model to write a new pattern

```timestamp
 14:27

```

formulate -> research -> criticize -> answer

```timestamp
 15:29

```

dewey decimal is indexing on a smaller set of data, compile all the relevant research and scale it