7 ms·

Agent GPT – Assemble, configure, and deploy autonomous AI Agents in the browser

98 points by kevinslin 3y ago

ed 3y ago
Wish it were possible to “deploy” an example agent without signing up. Even if the output were faked, it would help a lot to understand what this does. (I assume it’s “Geocities for AI agents” but couldn’t confirm.)
- byteknight 3y ago
  It's github page is here. Albeit with not too much more info. https://github.com/reworkd/AgentGPT https://github.com/reworkd/AgentGPT
  - kossTKR 3y ago
    Almost 23k stars!? Isn't that kind of crazy for a new niche project like this you've never heard of?
    - TeMPOraL 3y ago
      Stars are a weird thing. Some people seem to believe they're like likes on social media, or even that they indicate actual use. There's probably sites selling you fake stars, like they sell fake Instagram followers. However, many people, myself included, treat Github stars as just bookmarks, and star things they found interesting or possibly worth coming back to, regardless of the quality or whether one has any use for it.
    - capableweb 3y ago
      A lot of that is bought stars for sure, take a look at some of the profiles and you can tell immediately.
      - kossTKR 3y ago
        Interesting. So this combined with the fact that other comment say this is an incredibly thin wrapper on GPT-4 makes this some kind of investment bait? I wonder how profitable selling of non-products is with marketing like this.
- qup 3y ago
  I was mad about this too, but I'm interested enough that I was going to register. I use a custom domain and so would use something like "agent-gpt@customdomain.com" Anyway, they don't allow email signups, either. You have to use a social media login. So the pie gets even smaller. Edit: also the logo is not displaying properly on the signing page (Firefox, linux)
- ilaksh 3y ago
  There are a couple of challenges with making this stuff completely open like that right now. The first is that OpenAI has rate limits. They are especially small for GPT-4, which for anything complicated can be quite a lot better than 3.5. The other one is that if you do leave it open, then you can be sure that a significant portion of your customers will be from countries that could not pass the OpenAI phone verification. Or just didn't want to identify themselves. For some reason. Combine that with something that scares some people a lot like autonomous AI or connecting them to servers or the internet, and it feels like you might be on thin ice with OpenAI or some regulatory group. Especially if a bunch of Russian and Chinese users are finding you on some directory or post listed as a way to get around the phone verification. Just based on experience running my own service.
  - vidarh 3y ago
    Hence the "Even if the output were faked". Heck, even a video/screen recording would be a significant improvement.
SebaSeba 3y ago
At least as a non-native english speaker I was a bit confused as to what an agent would do compared to using plain ChatGPT. I tried then the examples "Plan a detailed trip to Hawaii" and "Write some code to make a platformer game". I tried the same Hawaii sentence with plain ChatGPT and told it that it can browse the web. Now I am thinking that AgentGPT does seem a good tool on top of ChatGPT. As they say on the Github page it "It will attempt to reach the goal by thinking of tasks to do, executing them, and learning from the results". This was a whole lot more thorough service than what plain ChatGPT did with the prompt. I'm just thinking that maybe they should emphasize more those points already on the app page, about the core value it adds and how it does it. Perhaps even explain it some more than what that sentence from their Github page does.
wilg 3y ago
This is a good line of thinking for making LLMs more powerful. I'm confused what I'm seeing though, can I not inspect the internals of what it's doing inside each task? And why does it take so long? Is it doing tons of GPT queries for each task?
- neuronexmachina 3y ago
  Looks like it's using LangChain internally, you can see the prompt templates here: https://github.com/reworkd/AgentGPT/blob/main/platform/reworkd_platform/web/api/agent/prompts.py https://github.com/reworkd/AgentGPT/blob/main/platform/rewor... And yeah, the initial "planning" query and the planned "tasks" are all separate GPT queries.
mercurialsolo 3y ago
Are agents not just a natural extension of ChatGPT as a product with chat being a universal language interface for the plugins which connect to tools and skills?
- bckr 3y ago
  Yes. In fact, ChatGPT with Plugins behaves like an agent. The idea behind these AutoGPTs is to give greater structure to the way the model thinks and acts, through a loop of planning, executing, and reflecting. Another goal is to provide the model with enhanced memory, similar to or identical to retrieval. In short AutoGPT-like applications treat the LLM as a reasoning core around which a scaffolding of higher-order thinking is built. It’s interesting, as a cognitive architecture. Imagine that your frontal cortex, responsible for planning, was very rigid and farmed out its actual reasoning tasks to your language-learning brain area.
theblazehen 3y ago
For something with more capabilities, check out https://github.com/Significant-Gravitas/Auto-GPT https://github.com/Significant-Gravitas/Auto-GPT I recently used it to have it find a source of data and scrape it as a test, and it works. Took only 30 min or so. Could have done it in under 5 min by myself, but taking 20s to write the prompt is still faster than spending the 5 min doing it myself
- johnnykree 3y ago
  i tried this some weeks ago and couldn't get it to work. I managed to implement the ChatGPT stuff but everything beyond that like implementing Google into it didn't work...
- almog 3y ago
  On what hardware did you run it?
  - capableweb 3y ago
    It's using GPT-4 (via OpenAI APIs) so anything that can make HTTP(s) requests should do the trick.
    - qup 3y ago
      640k oughta be enought for anybody (literally)
piyush_soni 3y ago
Hmm ... I wish with all these new agents my regular tasks become a bit easier, but so far it didn't work, it 'ran out of loops in the free plan'. Not going to pay without even executing one successful example (I know this one is a pretty tough problem, but I was willing to give some easier 'research' tasks too). My query was : "Find out the cheapest per unit Pampers New Born diapers from the ones available on amazon.in, flipkart.com, bigbasket.com, dmart.in, firstcry.com considering all the available card discounts, coupon codes, shipping and other charges etc. "
- SebaSeba 3y ago
  If you are a ChatGPT+ subscriber I think you can clone the repository and run the app for free locally with your own OpenAI API key.
  - piyush_soni 3y ago
    I'm not yet, because I haven't got confidence in ChatGPT's answers or abilities so far (e.g. the fact that it 'lies' with a straight face when it doesn't know the answer is especially troubling for me). I hope ChatGPT 5 or 6 are much better in this regard, and I will very gladly pay for it then.
    - pixl97 3y ago
      >e.g. the fact that it 'lies' with a straight face when it doesn't know the answer I see it's already ready to replace the sales team. Customer: "Can your product compute infinite loops in finite time?" SalesGPT: "Hell yes, our developers can do anything, would you like another vodka martini?"
  - danielbln 3y ago
    API access is not coupled with being a ChatGPT+ subscriber, and I believe GPT-4 API access is also not included in that. That means you can use GPT-3.5-turbo via API right now, and GPT-4 through API is available via waitlist.
- nologic01 3y ago
baxtr 3y ago
Can anyone explain what this does? I signed up and tried the Hawai example. Still not sure what it does and why it is relevant.
- reportgunner 3y ago
  > Still not sure what it does and why it is relevant I wonder the same about "AI".
smallest-number 3y ago
Projects like this try to get around the fundamental flaw in GPTs - namely that they do not have goals, plans, thought processes etc - without actually solving it, e.g. by having the AI write out its "goals" before continuing. But this is a hacky fix, and will never be reliable enough for consistent use. For that, more actual research is necessary, on how to simulate and model goals and trains of thought and have them interface with the world model provided by an LLM.
- wahnfrieden 3y ago
  The AI would never say hello, but if you say hello to it, it will say hello back. Is that also a hack? Aren’t you just describing everything about LLM behavior generally not only something specific about goals/tasks? In that case the nature of the thing is less interesting than the results we're able to find from it and I wouldn't worry about this kind of purity test.
- ianbicking 3y ago
  I feel like there's an implication here that the research should be in modeling architectures and training sets and other specialized machine learning. But there is research here: in natural language modeling of goals, plans, thoughts, processes, etc. Obviously we don't know what paths will be most successful. But a path where critical drivers of AI (like goals) are modeled in a transparent and comprehensible manner seems like a very attractive direction to take. I'd much rather be able to read my AI agents goals, plans, intermediate goals, self-analysis, etc., than have it all captured in a set of completely incomprehensible weights.
- pixl97 3y ago
  I mean most people don't have the resources needed to build a model big enough that these types of behaviors emerge so third party addons is all we got until Google/Microsoft/OAI drop something on us. Part of the issue here is the massive amount of compute needed over what we're already spending. ToT is showing a likely 10 to 20x number of calls to get an answer, which when you are compute limited is going to be a problem for deployment in mass. It's very likely we're going to have to wait for more/faster hardware.
m3kw9 3y ago
You know what I hear most? People asking if anyone had great success doing anything with these AutoGPT agents, it’s always meet with crickets. Either that or people don’t want to give the secret sauce
- dbecker 3y ago
  I've tried AutoGPT agents about 10 times for tasks that seem like they'd be a good fit. That meaning they need to scrape current data, but otherwise they'd be possible for a person willing to put in some time to combine data sources. An example is "create a table showing the current ratio of rental/sales prices (on a per square foot basis) for residential properties in the 10 most populated counties in Colorado." I'm still yet to get a reasonable result from trying this.
- relativeadv 3y ago
  In my own farting around with AutoGPT (which was quite a lot), what stuck out to me most was that the agents wouldn't rely on their own "knowledge" for tasks hardly ever. They would always default to browsing the internet for trivial questions and so it would take a good 10 minutes to arrive at a conclusion ChatGPT would've gotten in one-shot.
- _yb2s 3y ago
  You need gpt4 api access, and hardly anyone has been able to get that. Otherwise it will run on gpt 3, which isn’t able to function in this role. It’s hard to say how well it might work with very little info.
  - notadev 3y ago
    Hardly anyone has access to GPT-4? Isn’t it widely available now?
  - sureglymop 3y ago
    I've had that for months.. if you want that quickly you need a project in which you already incorporated the OpenAI API. At least that's what I did. I linked my open source project when applying and got access a few days later.
TonyBagODonuts 3y ago
We're sorry, you have run out of loops for the free plan. If you desire longer runs, consider subscribing at the bottom right. Shutting down. There's nothing in the bottom right.
vs4vijay 3y ago
You can find more info. about AutoGPT alike projects here - https://www.reddit.com/r/AutoGPT/ https://www.reddit.com/r/AutoGPT/

98 points by kevinslin 3y ago

ed 3y ago

Wish it were possible to “deploy” an example agent without signing up. Even if the output were faked, it would help a lot to understand what this does. (I assume it’s “Geocities for AI agents” but couldn’t confirm.)

byteknight 3y ago
It's github page is here. Albeit with not too much more info. https://github.com/reworkd/AgentGPT https://github.com/reworkd/AgentGPT
- kossTKR 3y ago
  Almost 23k stars!? Isn't that kind of crazy for a new niche project like this you've never heard of?
  - TeMPOraL 3y ago
    Stars are a weird thing. Some people seem to believe they're like likes on social media, or even that they indicate actual use. There's probably sites selling you fake stars, like they sell fake Instagram followers. However, many people, myself included, treat Github stars as just bookmarks, and star things they found interesting or possibly worth coming back to, regardless of the quality or whether one has any use for it.
  - capableweb 3y ago
    A lot of that is bought stars for sure, take a look at some of the profiles and you can tell immediately.
    - kossTKR 3y ago
      Interesting. So this combined with the fact that other comment say this is an incredibly thin wrapper on GPT-4 makes this some kind of investment bait? I wonder how profitable selling of non-products is with marketing like this.
qup 3y ago
I was mad about this too, but I'm interested enough that I was going to register. I use a custom domain and so would use something like "agent-gpt@customdomain.com" Anyway, they don't allow email signups, either. You have to use a social media login. So the pie gets even smaller. Edit: also the logo is not displaying properly on the signing page (Firefox, linux)
ilaksh 3y ago
There are a couple of challenges with making this stuff completely open like that right now. The first is that OpenAI has rate limits. They are especially small for GPT-4, which for anything complicated can be quite a lot better than 3.5. The other one is that if you do leave it open, then you can be sure that a significant portion of your customers will be from countries that could not pass the OpenAI phone verification. Or just didn't want to identify themselves. For some reason. Combine that with something that scares some people a lot like autonomous AI or connecting them to servers or the internet, and it feels like you might be on thin ice with OpenAI or some regulatory group. Especially if a bunch of Russian and Chinese users are finding you on some directory or post listed as a way to get around the phone verification. Just based on experience running my own service.
- vidarh 3y ago
  Hence the "Even if the output were faked". Heck, even a video/screen recording would be a significant improvement.

SebaSeba 3y ago

At least as a non-native english speaker I was a bit confused as to what an agent would do compared to using plain ChatGPT. I tried then the examples "Plan a detailed trip to Hawaii" and "Write some code to make a platformer game". I tried the same Hawaii sentence with plain ChatGPT and told it that it can browse the web. Now I am thinking that AgentGPT does seem a good tool on top of ChatGPT. As they say on the Github page it "It will attempt to reach the goal by thinking of tasks to do, executing them, and learning from the results". This was a whole lot more thorough service than what plain ChatGPT did with the prompt. I'm just thinking that maybe they should emphasize more those points already on the app page, about the core value it adds and how it does it. Perhaps even explain it some more than what that sentence from their Github page does.

wilg 3y ago

This is a good line of thinking for making LLMs more powerful. I'm confused what I'm seeing though, can I not inspect the internals of what it's doing inside each task? And why does it take so long? Is it doing tons of GPT queries for each task?

neuronexmachina 3y ago
Looks like it's using LangChain internally, you can see the prompt templates here: https://github.com/reworkd/AgentGPT/blob/main/platform/reworkd_platform/web/api/agent/prompts.py https://github.com/reworkd/AgentGPT/blob/main/platform/rewor... And yeah, the initial "planning" query and the planned "tasks" are all separate GPT queries.

mercurialsolo 3y ago

Are agents not just a natural extension of ChatGPT as a product with chat being a universal language interface for the plugins which connect to tools and skills?

bckr 3y ago
Yes. In fact, ChatGPT with Plugins behaves like an agent. The idea behind these AutoGPTs is to give greater structure to the way the model thinks and acts, through a loop of planning, executing, and reflecting. Another goal is to provide the model with enhanced memory, similar to or identical to retrieval. In short AutoGPT-like applications treat the LLM as a reasoning core around which a scaffolding of higher-order thinking is built. It’s interesting, as a cognitive architecture. Imagine that your frontal cortex, responsible for planning, was very rigid and farmed out its actual reasoning tasks to your language-learning brain area.

theblazehen 3y ago

For something with more capabilities, check out https://github.com/Significant-Gravitas/Auto-GPT https://github.com/Significant-Gravitas/Auto-GPT I recently used it to have it find a source of data and scrape it as a test, and it works. Took only 30 min or so. Could have done it in under 5 min by myself, but taking 20s to write the prompt is still faster than spending the 5 min doing it myself

johnnykree 3y ago
i tried this some weeks ago and couldn't get it to work. I managed to implement the ChatGPT stuff but everything beyond that like implementing Google into it didn't work...
almog 3y ago
On what hardware did you run it?
- capableweb 3y ago
  It's using GPT-4 (via OpenAI APIs) so anything that can make HTTP(s) requests should do the trick.
  - qup 3y ago
    640k oughta be enought for anybody (literally)

piyush_soni 3y ago

Hmm ... I wish with all these new agents my regular tasks become a bit easier, but so far it didn't work, it 'ran out of loops in the free plan'. Not going to pay without even executing one successful example (I know this one is a pretty tough problem, but I was willing to give some easier 'research' tasks too). My query was : "Find out the cheapest per unit Pampers New Born diapers from the ones available on amazon.in, flipkart.com, bigbasket.com, dmart.in, firstcry.com considering all the available card discounts, coupon codes, shipping and other charges etc. "

SebaSeba 3y ago
If you are a ChatGPT+ subscriber I think you can clone the repository and run the app for free locally with your own OpenAI API key.
- piyush_soni 3y ago
  I'm not yet, because I haven't got confidence in ChatGPT's answers or abilities so far (e.g. the fact that it 'lies' with a straight face when it doesn't know the answer is especially troubling for me). I hope ChatGPT 5 or 6 are much better in this regard, and I will very gladly pay for it then.
  - pixl97 3y ago
    >e.g. the fact that it 'lies' with a straight face when it doesn't know the answer I see it's already ready to replace the sales team. Customer: "Can your product compute infinite loops in finite time?" SalesGPT: "Hell yes, our developers can do anything, would you like another vodka martini?"
- danielbln 3y ago
  API access is not coupled with being a ChatGPT+ subscriber, and I believe GPT-4 API access is also not included in that. That means you can use GPT-3.5-turbo via API right now, and GPT-4 through API is available via waitlist.
nologic01 3y ago

baxtr 3y ago

Can anyone explain what this does? I signed up and tried the Hawai example. Still not sure what it does and why it is relevant.

reportgunner 3y ago
> Still not sure what it does and why it is relevant I wonder the same about "AI".

smallest-number 3y ago

Projects like this try to get around the fundamental flaw in GPTs - namely that they do not have goals, plans, thought processes etc - without actually solving it, e.g. by having the AI write out its "goals" before continuing. But this is a hacky fix, and will never be reliable enough for consistent use. For that, more actual research is necessary, on how to simulate and model goals and trains of thought and have them interface with the world model provided by an LLM.

wahnfrieden 3y ago
The AI would never say hello, but if you say hello to it, it will say hello back. Is that also a hack? Aren’t you just describing everything about LLM behavior generally not only something specific about goals/tasks? In that case the nature of the thing is less interesting than the results we're able to find from it and I wouldn't worry about this kind of purity test.
ianbicking 3y ago
I feel like there's an implication here that the research should be in modeling architectures and training sets and other specialized machine learning. But there is research here: in natural language modeling of goals, plans, thoughts, processes, etc. Obviously we don't know what paths will be most successful. But a path where critical drivers of AI (like goals) are modeled in a transparent and comprehensible manner seems like a very attractive direction to take. I'd much rather be able to read my AI agents goals, plans, intermediate goals, self-analysis, etc., than have it all captured in a set of completely incomprehensible weights.
pixl97 3y ago
I mean most people don't have the resources needed to build a model big enough that these types of behaviors emerge so third party addons is all we got until Google/Microsoft/OAI drop something on us. Part of the issue here is the massive amount of compute needed over what we're already spending. ToT is showing a likely 10 to 20x number of calls to get an answer, which when you are compute limited is going to be a problem for deployment in mass. It's very likely we're going to have to wait for more/faster hardware.

m3kw9 3y ago

You know what I hear most? People asking if anyone had great success doing anything with these AutoGPT agents, it’s always meet with crickets. Either that or people don’t want to give the secret sauce

dbecker 3y ago
I've tried AutoGPT agents about 10 times for tasks that seem like they'd be a good fit. That meaning they need to scrape current data, but otherwise they'd be possible for a person willing to put in some time to combine data sources. An example is "create a table showing the current ratio of rental/sales prices (on a per square foot basis) for residential properties in the 10 most populated counties in Colorado." I'm still yet to get a reasonable result from trying this.
relativeadv 3y ago
In my own farting around with AutoGPT (which was quite a lot), what stuck out to me most was that the agents wouldn't rely on their own "knowledge" for tasks hardly ever. They would always default to browsing the internet for trivial questions and so it would take a good 10 minutes to arrive at a conclusion ChatGPT would've gotten in one-shot.
_yb2s 3y ago
You need gpt4 api access, and hardly anyone has been able to get that. Otherwise it will run on gpt 3, which isn’t able to function in this role. It’s hard to say how well it might work with very little info.
- notadev 3y ago
  Hardly anyone has access to GPT-4? Isn’t it widely available now?
- sureglymop 3y ago
  I've had that for months.. if you want that quickly you need a project in which you already incorporated the OpenAI API. At least that's what I did. I linked my open source project when applying and got access a few days later.

TonyBagODonuts 3y ago

We're sorry, you have run out of loops for the free plan. If you desire longer runs, consider subscribing at the bottom right. Shutting down. There's nothing in the bottom right.

vs4vijay 3y ago

You can find more info. about AutoGPT alike projects here - https://www.reddit.com/r/AutoGPT/ https://www.reddit.com/r/AutoGPT/