Brian Stempeck
speaker
431 appearances
5 recordings
1 series
first heard Dec 2025
last heard 19 Dec
Brian Stempeck’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 5 in all, peaking in Dec 2025 with 5.
Appearances
So um we we drill into a few different areas.
So one good example is looking at ChatGPT, so the underlying large language model versus ChatGPT when it adds search.
Because so what what happens is there's the underlying LLM that the scientists there have trained.
And so you can use the API of ChatGPT and ask that questions.
And that'll just have the kind of the purest answer of asking the question just as a language model.
Then you can ask question to the more normal user interface that you or I would experience if we use the ChatGPT app or the website of saying, okay, what's the best SUV for family that goes skiing a lot?
And when you ask that question, the core model will sometimes supplement with a search of going into Google or Bing and adding search data.
And you can kind of see it doing this real time on Gemini, for example.
You'll see it running a search in the background, running fifty or sixty different queries, summarizing that information and adding it back.
But even on Chat GPT versus Chat GPT plus search, there can be meaningful differences.
There can be a twenty or thirty point swing of okay, the Honda C R V is being recommended fifty percent of the time versus being recommended eighty percent of the time.
And so that starts to indicate is you might be weak in the core model, but really good at SEO.
Or you might be strong in the core model, but weak it.
And so there's all these variations of what can happen of when you add search to the equation, does it make your standing worse or does it make it better?
And so even looking at just ChatGPT alone, you get different answers depending on are you asking the core model or are you asking the core model plus the search engine?
You zoom out from that a bit, the models have different opinions, right?
They have a different point of view about the best SUV from ChatGPT to Google to AI overviews inside of Google to Cloud from Anthropic to Deep Seek.
And so we we monitor all these different models because they swing drastically.
And you might say, well, why is that?
They're training on different data sets.
Showing 141–160 of 431 · page 8 of 22
← Previous
Next →