But if you look at projects like Autogen ( https://github.com/microsoft/autogen ), you see one master agent coordinating other agents which have a narrower scope. But you have to create the prompts for the agents yourself.

This GPTs setup will crowd-source a ton of data on creating agents which serve a single task. Then they can train a model that's very good at creating other agents. Altman nods to this immediately after the GPTs feature is shown, repeating that OpenAI does not train on API usage.

Prediction: next year's dev day, which Altman hints will make today's conference look "quaint" by comparison, will basically be an app built around the autogen concept, but which you can spin up using very simple prompts to complete very complex tasks. Probably on top of a mixture of GPT 4 & 5.

Also RIP chatgpt plugnins

The more elaborate the moderation tooling, the faster the race to the bottom gets as revenue drives black-hat marketers to become experts in circumventing AI moderation. It could end up being Google “S”EO all over again where we eventually ended up with very little real information in the results.

Moderation tooling is not as elaborate as we would hope, and all your work is with humans - who keep adapting.

But Actions will have all the same security challenges basically.

I'd guess this is more of an attempt to poach the market of companies like character.ai. The market for models with a distinct character/personality is absolutely massive right now (see: app store rankings) and users are willing to spend insane amounts of money on it (in part because of the "digital girlfriend" appeal).

The ban on "adult themes" is part of the reason people use services other than OpenAI for that kind of thing in the first place.

If so, I guess you can make such a "GPT" on your own server and independent from a specific LLM service by using a prompt like

    ...you have available an API "WEATHER_API". If you need the
    weather for a given city for a given day, please say
    WEATHER_API('berlin', '2022-11-24')
    and I will give you a new prompt including the data you
    asked for...

However creating something like this previously required a Jupyter notebook but now just...asking for it. Makes it accessible to 10x more people

The lack of transparency for how the product works behind the scenes will most likely make it difficult to build something effectively.

If you want reasoning capabilities that Open Source hasn't matched before today, and I'm guessing just got blown out of the water on again today... there's no reason to bother.

Or OpenAI's usage policy limits reasons (either because of your direct use, or because of the potential scope of use you want to support for downstream users of your GPT, etc.) Yes, OpenAI's model is the most powerful around. Yes, it would be foolish not to take advantage of it, if you can, in your custom tools that depend on some LLM. Depending on your use, it may not make sense to be fully and exclusively dependent on OpenAI, though.

If things change down the line the fact you invested 5 minutes into writing a prompt isn't going to be such a huge loss anyways, absolutely no reason to roll your own on this.

If things change down the road such that your tool (or a major potential downstream market for your tool) is outside of OpenAI's usage policies, the fact that you invested even a few developer-weeks into combining the existing open source tooling to support running your workload either against OpenAI's models or a toolchain consisting of one or more open source models (including a multimodal toolchain tied into image generation models, if that's your thing) with RAG, etc., is going to be a win.

If it doesn't, maybe its a wasted-effort loss, but there's lots of other ways it could be a win, too.

> If you want to be independent for academic/personal reasons, sure you can.

If your goal is to be in business, or to get some sort of reach, or anything other than "have fun tinkering"... wasting developer weeks on "could be a win" is how to fail.

What you describe are what’s called autonomous agents, and i agree that I also expected some interesting announcement there but probably too many security issues.

I saw that announcement and my immediate thought was "God yet another thing passive income youtubers will be shilling soon"

In general, I was a little confused by this. Sam's demo of creating a GPT didn't seem particularly exciting as well.

Plus, "generative pre-trained transformer" sounds futuristic, which seems like a fitting brand image for OpenAI.

https://en.wikipedia.org/wiki/The_China_Syndrome

I’d pay for one that was good at programming rubber ducking

There are specific sub-tasks that everyone would pay for to make their lives easier. This marketplace is trying to make that efficient

basically copying the Microsoft strategy of Embrace, Extend, Extinguish. Makes sense they took so much funding from Microsoft

The guarantee is that they will clone and extinguish most of the best bots/tools/services riding on top of GPT. It's what platforms overwhelmingly tend to do.

If your thing is a near-touch to GPT, depends on it, is of medium complexity or lower, and is very popular/useful, they will build your thing into GPT eventually.

On the flip side, if you're Salesforce and using GPT to augment something about a major product, you're not facing a serious threat of OpenAI trying to become the next big CRM company.

This process will work similarly to how it did with Windows, Google, AWS, etc.

The pattern will continue until no one needs anything. I am sure this company will eat everything. There will be no pie.

Also not only for ChatGPT, but there will be a timespan between majority of people are automated out of jobs and a a forceful redistribution of wealth happens (let’s call it socialism). It’s always a good idea to have that $$ to bridge that gap.

They might not. But if they do, I'd imagine there are a lot of people who will try. And as long as you're not dependent on the income stream they provide, you don't have much to worry about if it gets shut off.

It takes significant effort to come up with good use cases, build the prompts, and advertise the bots.

So a company can get a lot of value by going after this up and coming type of "content creator".

Right he said something about promoting the best but what about the discoverability?

People would pay for that.

Sh--... I better build it!

A good paper on the state of the Alexa skills BTW

SkillVet: Automated Traceability Analysis of Amazon Alexa Skills https://ieeexplore.ieee.org/document/9619970

Anyone with factual data (proprietary or not) is now an input away to AI / GPTs.

Data (or a new foundational model) is now the moat.

This makes me think that open source models are the future, for other reasons as well. No company is going to give them OpenAI their data to make bots, regardless if they won't train on the data. So what OpenAI can do at the moment is to focus on everyday user, and see where the biggest use cases are.

Furthermore, as Sam alluded to, this confirms to me that multi-LLM architectures (AutoGen, AutoGPT, BabyAGI etc) is the future. It´s about pushing the ways in which we can use these models, i.e. using different architectures because it will be a while before we get into more powerful models. These is simply no compute for it at the moment. The dust has yet to settle.

Edit: here is the message I get:

Your access to custom GPTs isn’t ready yet. We’re rolling this feature out over the coming days. Check back soon.

> Example GPTs are available today for ChatGPT Plus

or

> Starting today, no more hopping between models; everything you need is in one place.

Neither of which are true. I'm a paying user and I have access to neither. They do this _all the time_. They announce something "available immediately" and it trickles out a week or more later. If they want to do gradual rollouts (which is smart) then they should say as much.

The announced (and a few days ago leaked) Omni prompt also doesn't show up in the model selector. And that despite the expanded context looking very promising for the REPL-feedback-augmented code generation abilities.

New models and developer products - https://news.ycombinator.com/item?id=38166420

OpenAI DevDay, Opening Keynote Livestream [video] - https://news.ycombinator.com/item?id=38165090

Startups don't make sense as thin wrappers around another company's product when that company is aggressively expanding that product is a core offering that the vendor is aggressively working to provide as an integrated solution for as many markets as possible, which very much applies to OpenAI offerings in general, and its chat models especially.

A wrapper that also leverages some exclusive special sauce data, algorithm, etc., for which you have a real moat as a key component, that makes some sense. But just a thin wrapper around GPT? That's just asking to have your market eaten by OpenAI.

It might make some sense for products where the vendor is a stable, steady-state infrastructure supplier for many markets without any evident interest in entering the same market as the startup, where the uncertainty across markets that they would create by specifically targeting your startups market would hurt them more with their established customers than they would gain from your niche, but even that is risky because it requires lots of potentially-erroneous assessments of how the vendor would expect their other customers to react to them acting in your market.

I'd say that trying to build a business plugging gaps in OpenAI's infrastructure offerings (fine-tuning, RAG, etc.) is quite perilous, since this is really their core competency. But they've still had just one big user-facing hit so far: ChatGPT.

Wrapping OpenAI for a specific use case then building better workflows and UI for that use case than ChatGPT seems like it's still a pretty reasonable strategy. OpenAI isn't going to build the ideal workflow/UI for every use case.

It happened with hardware, operative systems, web tools such as maps etc.

You must use LLMs as a launching point to something else, some kind of 10x in a vertical.

Being able to "chat" is table stakes and worthless to you as a company by itself.

Jasper isn't having a good time, but you'd think the fact anyone can produce better output than they did after spending millions of dollars in GPT-3 based pipelines for $20 a month would mean they're dead dead.

But instead they went and changed their positioning, changed who their target market is, adjusted the UX, the messaging, and the feature set, and now it's a product that has a place even if OpenAI can give all of your marketers an internal ChatGPT (unless your plan is to have 100 different "GPTs" for every marketing task in your company)

tl;dr: People fail to realize that OpenAI can offer your startup's core value proposition tomorrow morning, and it doesn't matter if they don't offer it in a format that resonates with your target users.

You could have a cure to cancer and you'd still have to market it correctly.

Imagine if Jasper had raised and built their own GPT-3 alternative from the ground up at 100x the cost targeted at writing: it wouldn't have generalized like OpenAI's models given the different goals in training, they'd have spent orders of magnitude more, they'd currently be scrambling to partner with some non-OpenAI provider to play the role OpenAI does for them today...

Jasper "only" had to start integrating new APIs to keep up with the SOTA in quality, and instead of burning precious runway on R&D they get to focus on rebranding and driving home that positioning.

The point was that content generation use cases might be a viable path give OpenAI's thin-wrappers-killing direction.

Google launched Notebook LM[1] a while back which does a similar thing conceptually. It allows you to import a Google drive folder with docs of the stuff you would want to understand and then just chat with it. It's a good product but restrictive in the sense that it only allowed Google docs.

[1] https://blog.google/technology/ai/notebooklm-google-ai/

> Finally, ChatGPT Plus now includes fresh information up to April 2023. We’ve also heard your feedback about how the model picker is a pain. Starting today, no more hopping between models; everything you need is in one place. You can access DALL·E, browsing, and data analysis all without switching. You can also attach files to let ChatGPT search PDFs and other document types. Find us at chatgpt.com.

It's so annoying how they say this, I refresh and I still have to hop between models. Just say "rolling out over the next week" if that's what's happening. I even logged out and back in and still the same old way of doing it.

I do recommend logging off and logging back in if it isn't working. I've seen that update things for me.

It sounds like the GPT Store will solve some of the rough edges with plugins as they exist today, but I'm not interested.

OpenAI has been a horrible "partner" to work with. The team that owns this product is terribly under-resourced. You have no hope of reaching a human unless you have an inside contact. Support tickets are handled by GPT; policies are arbitrarily enforced; and major bugs in the API and documentation are simply ignored.

There's an amazing opportunity there for user acquisition, obviously, but do you really want to put your hand in the pain box?

how? login opens unrelated tab.

Found it:

https://chat.openai.com/gpts/discovery -> create own ->

https://chat.openai.com/gpts/editor

> Go to ChatGPT

Sad. Have a real-world use case ready to go and a plus account.

From the description of the past outside practice it is marketed as moving into OpenAI's offering, it sounds more like its custom prompts, not fine-tuned models.

This is the kind of feature that will prove to be a minor improvement for anyone on this forum, and a complete paradigm shift for the less technically-inclined. IMO.

1. RAG based infra building for people(regular joes).

2. Compartmentalize or Work around context lengths by introducing threads.

Then again, iOS felt like a very promising ecosystem when the iPhone 4 came out; I feel like it turned into an addendum/duplicate of the web rather than something genuinely original. Android seems more vivid, but not much more.

E.g. one popular comment in the submission about twitter’s new “edgy ai” was that it could be reimplemented as a chatgpt prompt[1]. Looks like this is even more relevant now.

[1]: https://news.ycombinator.com/item?id=38148845

Looking forward to seeing GPT4-Turbo there soon.

I think you'll find that "maximizing any and all productivity" is both harder than you think and not really something worth striving for.

Plus as neat as these advancements are, they are nowhere nearing the ability to create a "digital double of your brain" of yourself. There is a vast ocean from where we are are today to the possibility of replicating a brain digitally (if it's even physically possible).

It looks like we are moving away from apps to web GPTs, this looks like chatbot interface is here to stay and 'AI' is now the default interface that is to be expected.

I also don't need to spend lots of money on a developer to test my ideas out, this is great for product validation, I look forward to playing with GPTs.

I can see writing, travel and other bots being more enhanced and more powerful, hopefully existing startups will adapt to this change.

Exciting and interesting times!

For others like me that are getting errors accessing the page.

It doesn't seem to be much smarter than a simple "tell me how I should interpret inputs from the user."

C’mon OpenAI, we’re /trying/ to give you money here!

> We believe the most incredible GPTs will come from builders in the community. Whether you’re an educator, coach, or just someone who loves to build helpful tools, you don’t need to know coding to make one and share your expertise.

“Please work for us for free, while we keep all the product of your work for ourselves like we did with the content we scraped on the internet.”

OpenAI really is next level parasitism.

> Retrieval augments the Assistant with knowledge from outside its model, such as proprietary product information or documents provided by your users. Once a file is uploaded and passed to the Assistant, OpenAI will automatically chunk your documents, index and store the embeddings, and implement vector search to retrieve relevant content to answer user queries.

I'm so happy; I can finally ask questions about expo and trpc and get fresh answers. I confirmed this by asking chatgpt about the superbowl winners in 2022 & 2023.

This will most likely be like the google play store with a 99% of GPTs being a repackaged public prompt.

That said, your worry is one I definitely share. I guess I just hope more people think of ways they can try to ride/shape this wave, rather than stop/weather it.

On the long run, if your idea or app can be expressed as a flavor of a general GPT, you will not be able to compete with the AI gorillas. The space for AI startups is with custom, highly niched data or capabilities, that cannot be found in a general corpus, or that you can uniquely generate or control.

I have no doubt that this will immensely increase uptake among the less technically literate, since it will allow the techy people in their life (or on the app store) to introduce them with much less friction. It'll be like the little examples you can find on the "New Chat" screen of every chatbot, but 1000x more engaging