You get tighter feedback loops and better products if you own the vertical.
OpenAI already A/Bs test the responses it generates. Imagine if they own the text editor or spreadsheet you work on too. It’ll incorporate all of your edits to be self-correcting.
> Would love to hear OpenAI's explanation behind this line of thinking.
My assumption is that the data centers would mostly be staffed by those who will have to manually audit data going in and out of the LLMs on a daily basis. There would also be a need to generate/curate the test data that the LLMs will have train on.
There is potential for half a million jobs, but is that what you would want to have human effort invested in, is the real question.
One of the restuarant chains mentioned in the author's post (Social), is an extremely crowded pub during the night and for the rest of the time, a place where freelancers or remote workers come in to work and socialize. At least that was the case in Bengaluru,India before Covid.
I would say that from the restuarant's point of view, having the order-from-app experience works out since the freelancers can order via their laptops whenever they want, without having to flag down a waiter. And during rush hours, tables could order what they want without having to spot and call a waiter among a very drunk dancing crowd.
When I was junior/mid level engineer working in a 5-8 person engineering team, there was this particular senior engineer whom I disdained. It had nothing to do with them as a person, but I always had the feeling that they just did things at work without caring for the impact of it.
The team and company grew in size over a period of 2 years and I went through a lot of first hand experiences of having to deal with unrealistic deadlines, chaotic lines of communication and multiple nights of tailing production logs to fix bugs. It was around this time, that the senior engineer during a company even got drunk and talked to me about their first few jobs.I realized that what they described, was quite similar to what I was facing right now and I came to understand why they cared so little about work they do on a daily basis.It honestly scared me at to think that I might either end up like this person or worse or burning out and quitting the field.
Right now, I am close to age the senior engineer was when I met them and even though I may selectively decide not to involve myself with things that can overstretch me, I am in no way the same as that other person.
The reason I am sharing this tangential story is to highlight my opinion about old age. It does not matter much if you have gained a lot of experience as you grow older, unless you are able to use it effectively.
To anyone in a similar situation; you might want to spend some time reflecting on the reality of your role - one of the failure modes of incompetence is using "care" as a proxy for "ability" because an incompetent person almost by definition can't assess ability directly. I fell for that once and consider myself lucky that I figured out what I was doing wrong while still a fresh-faced graduate.
The ideal engineer is not emotionally involved in their work. They aren't shareholders; they don't reap the benefits of what they do, they don't control the direction of the company. They are there to achieve specific technical goals with professionalism. Bringing emotion or non-technical factors in to the deal is an obstacle to excellence.
The ideal engineer is not emotionally involved in their work.
As much as some may profess to actually believe that it simply isn’t even close to true.
I am sure there are some unicorn exceptions (like the Spock character from Star Trek) but those who are excellent at their work also put a lot of time and effort into it, if not continuously then they’ve at least paid their dues for a span of decades. However they got there it represents a significant emotional investment involving pride and care or even a kind of love.
One can’t know everything all the time. There are ALWAYS going to be gaps in your own ability especially when you’re doing new and difficult work, and the way to bridge those gaps is to be driven by a desire to overcome them— in other words by CARING enough to keep trying.
The most successful people I can think of personally always care enough to overcome gaps in their ability. It looks like incompetence at first, then it becomes obsession, then it becomes problem solving, and it is ends with mastery, IF they CARE enough.
Maybe. But if someone is a junior employee they aren't going to be able to identify all the specific tells for that. I know a passionate engineer who's meeting presence can often be mistaken for someone that is half asleep. He doesn't have a lot of respect for meetings and doesn't engage much. Would a junior engineer in that meeting identify that his dozey lids, slack jaw and mastery of simple programming conceal a burning passion for the art of great software? I'd bet not. You need to be quite good at programming to properly assess what is going on.
I disagree that blindly executing like a robot is excellent. A truly excellent product happens when people go above and beyond, and think of the little yet related things, instead of doing everything literally and exactly as told and not giving a shit about the end result.
Sounds like he might have been acting on wisdom, and at the time you might have perceived him through a lense of naivete, which still holds. I felt similarly in my early twenties in perhaps a similar environment, but now I know all the seniors I'd have resented were decades ahead of me in terms of what they valued and why. Caring about that work was stupid, and I should have just enjoyed my time more.
Their decisions made the life of those around them miserable because they did not care about the consequences of it. Like for example, hiring an intern who had basic programming skills when there was no engineering bandwidth available to support the intern, and then bouncing them around multiple large projects every week, for the sake of showing upper management that they had people "working" on priority tickets. There is more to this story, but it's just more details painting that person in a bad light.
I, most certainly am not such a person.
> Caring about that work was stupid, and I should have just enjoyed my time more.
I agree. It's a work in progress for me to get to that state.
I do not believe that most of these `perfectionists` are trolls. Some have just very bad experiences either in their career or childhood that make them feel that making mistakes is not normal.
I was a `perfectionist` for a while due to certain bad experiences at work, and it was only through the help of really good teammates that I was able to slowly get rid of it. And that required pointing out things like what the author has done in their blog post and once someone sees that mistakes are things that anyone could make, they get more comfortable the concept.
The harder part is understanding why a perfectionist is so, and then not getting frustrated while you try to help them improve.
The screenshot of the tweet thread at the end seemed to indicate that the alignment team wanted to explore other avenues(or models) to get to AGI, while the company leadership wanted to keep improving on the transformer models.
From the business point of view, continuing to improve the goose that is laying the golden eggs makes sense, but as a researcher they might have already ended up seeing the limit of adding compute and data, and wanted to pivot. As someone who worked a mid level manager/engineer, I can understand the frustration involved in making the management level understand what's really happening on the ground level.
From my point of view the underlying problem here seems to be that someone in Quora product team feels that the best way to drive engagement is to predict what the user might want to know about and then spam them with it, rather than letting them discover on their own or use their feedback.
I stopped using Quora 8 or 9 years back, because despite how much time I spent curating the feed I was served, the questions always seemed to veer back into the questions on life experiences, `write a short story in 3 lines` or about relationships. This was most probably due to topics like these being a fad in India during the time and often received a lot of answers or upvotes.
Every time I marked an question as something I was not interested in, because of the topic, Quora assumed that I did not like the author's answer and then proceeded to show 5 other questions the author has responded to. I assumed it was because of some bug on their side ignoring my preferences of topics, until I spent a week in Dubai and saw most of the questions in my feed match the topics I marked as interested.
> Additionally, members of the program receive priority placement and “richer brand expression” in chat conversations, and their content benefits from more prominent link treatments. Finally, through PPP, OpenAI also offers licensed financial terms to publishers.
> A recent model from The Atlantic found that if a search engine like Google were to integrate AI into search, it would answer a user’s query 75% of the time without requiring a clickthrough to its website.
If the user searching for the information finds what they want in ChatGPT's response (now that they have direct access to the publisher data), why would they visit the publisher website ? I expect the quality of responses to degrade to the point where GPT behaves more like a search engine than a transformer, so that the publishers also get the clicks they want.
Average user will NOT CLICK on those links. Anyone who ever had a news site and did some research how people interact with the content knows this. You show the source, but only a tiny amount of users click on those links.
I believe what the parent comment is trying to imply is that the search results are fetched/retrieved from Bing's own internal ranking vector(?) database and then passed to the LLM, which then converts the received documents into a more human readable format and fills in any missing gaps in the information with it's own data.
So the gaps are the only areas where the LLM can hallucinate on and if your search query is easily available information on the internet, then hallucinations will be less or none.
Edit: I have used RAG with a project that I am working on and it's quite hard to ascertain if the LLM used the information provided as part of the RAG documents or just made up information on it's own, since even without RAG, we were getting similar responses 7 times out of 10.
Based on the way you have described their API, I can make 2 assumptions
1. The might have had well supported API until the point at which they realized they wanted to have more people to use the webapp, rather than having the data sent through an API. More time spent on the Hubspot webapp means, more chances the customer notices the other services Hubspot provides and pays for them.
2. The versioning APIs that you see are most likely due to an enterprise deal Hubspot made with a large company who asked for variations in the API, and Hubspot introduced a v2/v3/v4 instead of modifying v1 (to not cause disruptions for their customers using v1).
