Hacker News Reader

Comment by skilled | Hacker News Reader

> when did lebron james reach 36,000 career points

> As of my last update in January 2022, LeBron James had not reached 36,000 career points in the NBA. However, I don't have data on games or events beyond that date. You may want to consult the latest NBA statistics or relevant websites for up-to-date information on LeBron James' career point total.

https://edition.cnn.com/2021/12/29/sport/lebron-james-steph-...

View on HN

My cutoff date is in September 2023 but I didn’t know this one either.

Same... I just looked up stories from that date.

How can you be sure they weren't hallucinations?

They are not, as per:

https://chat.openai.com/share/e4d1e5e3-f603-4d66-b11c-4c1a93...

(taken from this article: https://stackdiary.com/chatgpts-cutoff-date-upgraded-to-janu...)

> The 2021 Formula 1 World Championship was won by Max Verstappen of Red Bull Racing. It was his first world championship title, and he secured it after a closely contested battle with Lewis Hamilton throughout the season, culminating in a dramatic final race at the Abu Dhabi Grand Prix.

They've got a lot of data in to be able to identify that:

- first championship

- closely contested battle

- dramatic final race

all are true.

Right, how do you know this isn't a Gettier problem?

I'm not sure what you mean or why you're trying to refute my reply but I'm not really interested in this kind of discussion. I've presented both "facts" in various different comments in this thread and I couldn't care less if the data was upgraded to be from 2023. I'm sorry if I left an impression that I do.

We live in a society here! ;)

I have January 2022 for GPT-4 and September 2021 for GPT-3.5

I think you mean September 2021 because that's what it has been on for a long time (and that's what mine is currently saying).

You should take some time to learn more then. Some interesting things have happened in the past two years.

Parent comment is talking about their brain

Based on the article you’ve linked, which is from Dec 29th 2021, this answer is wrong.

Looks like chatgpt not having enough data to provide the correct answer, or chatgpt giving wrong answer despite having access to relevant data.

Some other experiments,

> when did Siobhan Cattigan pass away

> Siobhan Cattigan, the Scottish rugby player, passed away on November 20, 2021. Her untimely death was a shock to the rugby community, and she was remembered for her contributions to the sport and her vibrant spirit on and off the field.

https://chat.openai.com/share/4fc16baf-7fd5-42f1-b734-b0d124...

https://edition.cnn.com/2021/12/01/sport/siobhan-cattigan-di... (actual date of death is November 26)

> when did Claude Humphrey pass away

> As of my last training data in January 2022, Claude Humphrey, the former NFL defensive end primarily known for his time with the Atlanta Falcons and Philadelphia Eagles, had not been reported to have passed away. To get the most up-to-date information, I'd recommend checking the latest news sources or online databases.

https://chat.openai.com/share/0444df5b-96d6-4242-9253-e41acd...

https://edition.cnn.com/2021/12/04/sport/claude-humphrey-pro...

Wow, a lot of people have died recently.

a lot of people die all the time, they just become more relevant to you with time.

I am waiting for it to get a better beyond plugins and have some ability to query an api for more detailed facts.

A friend told me he uses Bard for current data. I have not tried it, so I can’t confirm.

I'm hoping that one day ChatGTP will be advanced enough to ignore advertisements for me. Oh what will the world do if they have to go see a doctor to get an effective toenail fungus cure rather than try the four hundred clickbait remedies that don't.

I use bard for current data. I can confirm that it's up to date and will search the live internet

Going to reply to myself here to say that it does seem like there is definitely information from December/November 2021 available, like clear-cut facts.

I'm just curious why OpenAI didn't make an announcement or what's the deal here, wouldn't this warrant retraining the entire model?

They probably won't share how they did it, but there's been a lot of research over the past 6 months showing how you don't have to retrain the entire model to add in new sources. I know nothing about this stuff, but my limited understanding from blog posts is it's easier than anyone had thought to add in new data to a pre-existing model.

Do you by any chance have any of these blog posts available for my own reference? If not you maybe someone else does, I don't recall seeing it but it sounds interesting.

I assume they mean fine tuning?

https://en.wikipedia.org/wiki/Fine-tuning_(deep_learning)

https://platform.openai.com/docs/guides/fine-tuning

I think there was a paper from Google showing that if you included 5% of your original dataset together with the new data during the finetuning then catastrophical forgetting didn't occur. Perhaps it's that simple.

they are working on including information up to 2025. now THAT will be a big announcement.

Since gpt will be writing all the text, this is possible.

This is so true.

Chatgpt model is constantly training and it doesn’t require retraining the whole model.

Can you provide specifics?

Only that openai has said they are constantly changing the chatgpt model.

Right...

Do you really want it to confidently answer that question based on a single article?

Of course I do because it’s not a single article. It was covered by at least 500 different publications and social media sites.

I am curious as to why you made that comment though… you can just go and verify it for yourself. It does seem to have data up to December 2021 now.

It’s just not as broad as you’d expect from an actual new model. That is the interesting part and that is what I pointed out.

mine is still 2021

> As of my last knowledge update in September 2021

This bit about knowledge cutoff actually is part of the internal pre-prompt which hardcodes September 2021 as a constant (i.e. the service before the AI model itself internally adds "knowledge cutoff: September 2021" to every prompt). So it doesn't say anything about what the real cutoff is. It was discovered months ago when someone managed to make ChatGPT leak the original prompt (and it makes sense: how would the model know about its cutoff otherwise? It's not in the training data). Also we can't tell if it's hallucinating or not when mentioning a different date.

Reply on news.ycombinator.com