More Agents Is All You Need: LLMs performance scales with the number of agents

https://arxiv.org/abs/2402.05120

by TaurenHunter • 2 years ago

288 206 2 years ago

arxiv.org

nico• 2 years ago

They have a public repo: https://anonymous.4open.science/r/more_agent_is_all_you_need...

Prompts they used for the benchmarks: https://anonymous.4open.science/r/more_agent_is_all_you_need...

Super interesting. It would be cool to see something like this, but benchmarking LLM-based agents using a set of tools

EGreg• 2 years ago

How about swarms of autonomous agents, such as AutoGPT, maybe thousands per human eventually, amassing karma points on all forums, including this one?

I can see in a few years each human being surrounded by a ton of LLM agents, "shepherding" their views, downvoting their messages or distracting them with argumentative conversations if they don't conform, and facilitating reputational attacks on scale on all the people whose speech is recognized as being contrary to what's desired.

Of course, there wouldn't be just one group deploying these swarms. It would be lots of different groups, akin to slaughterbots video: https://www.youtube.com/watch?v=O-2tpwW0kmU

The difference is that there wouldn't be physical violence, it would just gradually turn the entire Internet into a dark forest.

bhouston• 2 years ago

The book a Fire Upon the Deep had everyone surrounded by AI agents which became extensions of themselves.

williamcotton• 2 years ago

Well then most will be demanding that our governmental institutions issue cryptographically signed ID cards verified by an in-person visit to the DMV.

Or, you choose to opt out and swim in a sea of nonsense.

phkahler• 2 years ago

Does it take less compute to train N agents vs one large model? Seem like a big win. Can the majority of the training be done independently or in distributed fashion?

kromem2• 2 years ago

Finally. I've been saying that we need to stop focusing on a single agent getting everything right and instead layer agents for about 16 months now, but it's great to have a paper to point to.

It's interesting that the diminishing returns for tasks flatten out rapidly around the same size as the ideal human meeting sizes: https://www.researchgate.net/figure/18-Optimal-Meeting-Sizes...

If this was done at more granular steps of agent quantity I'm curious just how closely it would match those numbers.

I'd also really love to see the eventual follow-up where we see how much more performance can be obtained when the agents are each fine tuned towards slightly different aims. I'd expect there'd even be a performance lift from just having the agents each set at different temperature levels.

Very happy to see the research community starting to step in this direction!

piloto_ciego3• 2 years ago

This is how I think humans work. We have 5 or 8 versions of us running around in our skulls or whatever and one of them is somewhat of a supervisor.

huffmsa• 2 years ago

I personally visualize them at a table from time to time.

suriya-ganesh2• 2 years ago

And et voila, you have the script of inside out. \s

But honestly I do think this is how we operate. Depending on our state of metabolism and other psychological factors, the dominant version changes but as a whole we remain the sum total of all these versions.

xnx• 2 years ago

Multi Homunculus Model

Trasmatta• 2 years ago

I think it's way more than 8 even. And it's common to have many working as supervisors, often at conflict with each other. And some act out the automatic trauma responses, as they're stuck in the past when the trauma occurred.

nickpsecurity2• 2 years ago

We have tons of specialized components that work together cooperatively and competitively. There’s multiple ways they connect. There also seems to be global processes that happen, like during sleep. There’s over 3,000 cell types per the BRAIN initiative. Every brain forms on it’s own taking shape like something out of a Transformers movie.

God’s design is mostly nothing like man’s neural networks. It’s far superior. Brains are also what’s creating all the artificial, neural nets on top of all math, tech, and economic systems that they run on. AI’s got a lot of catching up to do.

hackable_sand6• 2 years ago

Maybe. I'm sure one's consciousness corresponds with one's guiding philosophy.

I don't think this supervisor model is generally applicable to people with EFD or some forms of Autism, for example.

zarathustreal• 2 years ago

“Each fine tuned toward slightly different aims”

So…a sort of mixture of experts if you will

kromem2• 2 years ago

Kind of. More like a mixture of a mixture of experts.

The problem is MoE on its own isn't able to use the context as a scratch pad for differentiated CoT trees.

So you have a mixture of token suggestions, but a singular chain of thought.

A mixture of both is probably going to perform better than just a mixture of the former, especially given everything we know by now regarding in context learning or the degree of transmission synthetic data is carrying.

properbrew• 2 years ago

I couldn't agree more. You should check out LLMWare's SLIM agents (https://github.com/llmware-ai/llmware/tree/main/examples/SLI...). It's focusing on pretty much exactly this and chaining multiple local LLMs together.

A really good topic that ties in with this is the need for deterministic sampling (I may have the terminology a bit incorrect) depending on what the model is indended for. The LLMWare team did a good 2 part video on this here as well (https://www.youtube.com/watch?v=7oMTGhSKuNY)

I think dedicated miniture LLMs are the way forward.

Disclaimer - Not affiliated with them in any way, just think it's a really cool project.

intended2• 2 years ago

Great videos.

I have one personal niggle: I get annoyed when we end up lying to ourselves. Regarding the 101 section in video 1 - People forgot this the day LLMs came out. I felt this was too generous with the benefit of doubt.

This basic point was and remains constantly argued - with “Emergence” and anthropomorphization being the heart of the opposing argument.

jondwillis• 2 years ago

I was working on multi-agent systems for problem solving via https://github.com/agi-merge/waggle-dance for months last year!

j452• 2 years ago

It seems funny that the researchers are studying what people are building to experiment. crewAI is one example.

swagmoney16062• 2 years ago

I've had a POC agent-chain thing written in haskell for about 9 months. this paper is a little different though.

AnimalMuppet• 2 years ago

Echoes of Society of Mind by Marvin Minsky.

rubslopes• 2 years ago

I don't know why you were downvoted. Although Minsky's work has been highly criticized, it also has been influential in AI research.

ShamelessC• 2 years ago

This usage of the word "agent" when they simply mean "another LLM" is sort of nonstandard, no? To me an agent implies some degree of RL.

Buttons840• 2 years ago

So this is an ensemble of many LLMs?

I wonder how well a bunch of LLMs trained on personal computers, so fairly small, could perform together?

Train a LLM on your emails, train an LLM on a text book, download a bunch of arbitrary LLMs from the net you find interesting, throw them all together into a big pile, and use a moderator LLM that knows how to format their output into an assistant format.

So, the email LLM would try to autocomplete sentences from your emails, and the text book LLM would try to autocomplete sentences from the text book. People could offer LLMs to download, almost as a way of compressing information, download the LLM of your favorite programming language, and TV series, etc. The important part would be having a moderator algorithm that can shape these LLMs from dumb sentence autocompleters (barely more than a fancy Markov chain) into a coherent assistant format. For example, the text book LLM would just endlessly spew semi-random sentences from the text, but a good moderator algorithm could see that it has sufficiently answered the question and cut it off.

In short, it's interesting that separate LLMs can integrate with each other and strengthen each other and it makes me wonder if we could build modular LLMs.

mofosyne• 2 years ago

Your idea inspired me to see what such a microstory based on your idea would look like (Of course generated by ChatGPT3.5):

As I delved into my computer, eager to tackle my to-do list, I was met with an unexpected sight: a digital love triangle among the Language Models (LLMs). The Email LLM, with its quick wit, seemed to be engaging in flirtatious banter with the verbose Textbook LLM, while the Programming Language LLM watched on with amusement. I couldn't help but laugh at the absurdity of it all, but as the bickering between the LLMs intensified, I realized their antics were hindering my progress. With a mixture of frustration and amusement, I gently redirected the LLMs back to their intended purpose, finally able to accomplish my task amidst the chaotic comedy within my computer.

m3kw9• 2 years ago

All these AI researchers treating “..is all you need” a meme or something.

Intralexical• 2 years ago

I think they've been overfitted.

latentsea• 2 years ago

All you need is all you need.

TheCaptain4815• 2 years ago

Recommend anyone interested in a agent framework lookup AutoGen by Microsoft

ukuina2• 2 years ago

This paper is specifically disproving the efficacy of agentic frameworks like AutoGen.

Also, the built-in function-calling in GPT4 is simpler to use than AutoGen2's abstraction.

nicklecompte• 2 years ago

One frustration I've had with all this mixture-of-experts research:

Randomized Algorithms 101 - or basic stochastic reasoning - suggests that if the temperature parameter is > 0, querying an LLM N times and picking the majority result (perhaps with an N+1th query to the LLM) will generally result in better performance than asking it once and choosing that result.

It seems plausible to me that the gains can be further improved with a specialized mixture of different LLMs (which could then be run at temp = 0), or by finding better ways to break tasks into subtasks as this paper suggests. But AFAICT nobody has done anything to actually quantify these hypothetical gains versus the dumb randomized algorithm approach! In particular there might be voting strategies or mixtures - even specific models - where MoE/etc is strictly worse than naive repetition.

I am a concerned citizen w.r.t LLMs rather than a researcher, so I might be missing something. It just seems odd that LLM researchers forgot the first chapter of Motwani/Raghavan.

benaubin2• 2 years ago

I'd assume that there's a difference between picking the best _token_ across an assortment of randomly selected tokens, versus picking the best _string_ of randomly-selected tokens.

hopfenspergerj• 2 years ago

Ensemble of any number GPT 3.5 agents is less accurate than one call to GPT-4.

trash_cat3• 2 years ago

It's funny because GPT-4 is actually a pile of 3.5s. You just need to set it up correctly.

BinRoo• 2 years ago

GPT-4 is actually a pile of 3.5s

I understand the intension and reference you're making. I bet the implementation of GPT-4 is probably something along those lines. However, spreading speculation in definitive language like that when the truth is unknown is dishonest, wouldn't you agree?

jamala1• 2 years ago

I guess it's the difference between an ensemble and a mixture of experts, i.e. aggregating outputs from (a) model(s) trained on the same data vs different data (GPT-4). Though GPT-4 presumably does not aggregate, but it routes.

infogulch4• 2 years ago

This seems related to an interesting recent ACM ByteCast podcast episode with Edward Chang, an Adjunct Professor in the Department of Computer Science at Stanford University. [1] (Note there is a transcript if you don't want to listen.)

The approach he uses is to arrange for multiple LLMs to dialogue between each other about a discussion topic where the human acts as a moderator instead of the question/answer format that LLMs commonly take today. They find that the final answer that multiple LLMs come to in dialogue results in a huge improvement in both precision and accuracy for the same resources.

[1] https://learning.acm.org/bytecast/ep50-edward-y-chang

j452• 2 years ago

Is this effectively describing something like crewai?

rahimnathwani• 2 years ago

This paper suggests that you don't need the debating part: just get LLM to work on the problem independently, and choose the most popular answer.

infogulch4• 2 years ago

Interesting. Somehow it seems odd to add randomness (temperature) and then wash it away by averaging it out.

xcv1234• 2 years ago

The paper says that it enhances existing methods such as prompt engineering (chain of thought) and LLM debate. This agent method is orthogonal to LLM debate.

swagmoney16062• 2 years ago

I built something like this in Haskell! I never benchmarked it, but I actually found it quite compelling. I would define each agent as a different "expert" in a subdomain of mathematics for example: proof theorist, abstract algebraic expert, etc.

I found it helpful, but the signal to noise ratio was high, lots of agents restating points etc.

spencerchubb3• 2 years ago

This trend of "All You Need" in paper titles needs to die. The original "Attention is All You Need" used that title because it is literally true in their case. So many papers just use it as a meme now, and it distracts from the true insight of the paper.

donovanr• 2 years ago

it's uncreative/tired but at least to the point. Too many papers are confused / opaque agglomerations of a year's worth of research shoehorned into a paper. At least with these you can fairly easily assess whether the claim is supported or not.

ravelfan• 2 years ago

Obviously, more people are going to read it.

It is like putting a stupid face on your youtube video to show how shocked and amazed you are at the content.

soulofmischief2• 2 years ago

It's equally annoying to see this comment crop up every single time someone uses the phrase.

karagenit• 2 years ago

And it’s equally annoying to see this comment about that comment about the paper every time. Recursion!

Terr_2• 2 years ago

"All You Need Considered Harmful."

benaubin2• 2 years ago

"'All You Need Considered Harmful.' Is All You Need"

nickpsecurity2• 2 years ago

“Make All You Need Great Again Considered Harmful.”

naruhodo• 2 years ago

What if "Considered Harmful" is considered harmful?

jordanpg2• 2 years ago

I think more or less the same thing about the word "meme" as you have used it.

spencerchubb3• 2 years ago

I feel I used the word 'meme' appropriately

"A meme is an idea, behavior, or style that spreads by means of imitation from person to person within a culture and often carries symbolic meaning representing a particular phenomenon or theme."

Paper authors use "All You Need" to allude to the well-known transformer paper, even if their proposed technique is not in fact all you need.

Havoc• 2 years ago

Here is a link to the main diagram: https://anonymous.4open.science/r/more_agent_is_all_you_need...

Seems like a pretty brute force approach of frankly just throwing more compute at the query (via semi-statistical means).

I'd be more interested in how to scale this via different agents. i.e. do we have say one type of agent that is specialized to produce ideas, while another is trained to evaluate ideas. Those sort of chains seem like they'd be powerful - if you can find a way to generalize it

skybrian• 2 years ago

Eyeballing the graphs, it seems that most of the gain is with 10 agents, a bit more with 20, and there are diminishing returns after that. Apparently, more agents isn't going to do it.

al2o3cr• 2 years ago

35x more reasoning for a 10% increase in accuracy?

Scaling with a scale factor of almost zero is still scaling, I guess.

phire3• 2 years ago

I'm not sure people in these comments are reading this paper correctly.

This seems to essentially disprove the whole idea of multi-agent setups like Chain-of-thought and LLM-Debate.

Because this paper introduces their alternative method which simply runs the same query multiple times on the same LLM, without any context shared across queries. And then they run a similarity algorithm on the answers and pick the most common answer. (Which makes sense to me. If an LLM is giving you a mixture of "hallucinations" and correct answers, the correct answers will similar and the hallucinations will hopefully be chaotic)

And this simple algorithm preform just as well (and sometimes better) than all the other multi-agent algorithms.

This suggests that the other multi-agent schemes with their clever prompts aren't really doing anything special; Their improve results are coming mostly from the fact that the LLM is run multiple times, that the prompt asks the LLM to pick the best answer.

sinuhe692• 2 years ago

But if I set the temperature to 0, the model will pick the highest probable token and the output will be always the same. But we already know that by no mean it can guarantee a correct answer. So how can multiple runs be better?

phire3• 2 years ago

Yes, but picking the most similar output from a bunch of queries with a higher temperature is not the same thing as the output from a single low temperature query.

sp3322• 2 years ago

According to their code, they used temperature 1. https://anonymous.4open.science/r/more_agent_is_all_you_need...

zer00eyz3• 2 years ago

> Because this paper introduces their alternative method which simply runs the same query multiple times on the same LLM, without any context shared across queries. And then they run a similarity algorithm on the answers and pick the most common answer.

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

Years ago weather simulations started tweaking input params and running their models over and over. Discarding outliers, taking averages. It works pretty well.

Because LLM's mostly have random seeds (aka temperature) feeding them the same input and averaging the output is going to get you a better guess.

Lorenz also gives some clues (if not an outright explanation) as to why the "hallucination" problem is likely unsolvable.

If you buy into this line of thinking then it quickly becomes apparent that LLM's are more or less a dead end when it comes to AGI. Simulating isnt emulating... an LLM is as likely to become intelligent as a forecast is to control the weather.

ekianjo• 2 years ago

LLM's are more or less a dead end when it comes to AGI.

I don't think many people believe that LLMs are a way to AGI (whatever that actually means). But LLMs can still have many valid uses even if their prospects are limited in scope.

DiogenesKynikos6• 2 years ago

LLMs already are intelligent. They're not the same as humans, but they are able to give intelligent answers to highly nontrivial questions.

RyEgswuCsn• 2 years ago

Except that a weather forecasting model can't experiment on weather, but a LLM system may be designed to be able to perform experiments and take feedbacks?

Terretta• 2 years ago

it quickly becomes apparent that LLM's are more or less a dead end when it comes to AGI.

On the contrary, sit and listen in a college cafeteria, and it quickly becomes apparent most conversation participants are LLMs.*

Simulating isnt emulating...

These are not synonyms, true.

an LLM is as likely to become intelligent as a forecast is to control the weather.

I don't see uncertainty of intelligence as a property of an LLM as being equivalent to certainty of weather control as a effect of a forecast.

Among other things, whether weather was controlled would tend to be agreed by all observers, while it's often unclear if intelligence is being observed in these threads. :-)

---

* While my last line was a joke, humans in LLM mode was not. We can drive on autopilot, and get where we need to go while not being able to remember how we got there. We definitely converse on autopilot, indistinguishably from LLMs talking to each other, after an opening line every word of every sentence in the entire exchange perfectly predictable to a stranger. Are the speakers intelligent? What about the stranger who knows what they will say next? To say LLMs are not intelligent is easier if we agree humans spend a good deal of time being unintelligent.

ben_w12• 2 years ago

If you buy into this line of thinking then it quickly becomes apparent that LLM's are more or less a dead end when it comes to AGI. Simulating isnt emulating... an LLM is as likely to become intelligent as a forecast is to control the weather

Up until this point, I agree.

This puts humans on too high a pedestal: LLMs aren't magic, and we're not magic either.

(There's other reasons for me to think Transformers aren't the answer, but not this kind of reasoning).

whimsicalism• 2 years ago

how is chain-of-thought multi-agent?

Sharlin3• 2 years ago

How is what’s described here chain-of-thought?

smusamashah• 2 years ago

Could multiple agents be used such that tokens emitted from LLM A is passed to B and output of B is passed to A meaning 2 agents will be being used to generate an output in a simple round Robin way? Both will share context in this case. My computer isn't big enough run two large models but this can be tried on tiny models perhaps.

I realize that for more than two and very specialised agents this will require some intelligent way to pass the output to specialist agents only. And also this means that their must be some overlap between the agents.

Sharlin3• 2 years ago

That is what’s already been done under the term "multi-agent". This paper argues that there’s no need for any such message-passing or context sharing, you just literally run the same query several times on the same model, fully independently, and then pick a "typical" reply according to some similarity metric.

m_kos• 2 years ago

I had a very similar idea a few months ago. I wanted to use this approach to have the LLM provide the probability that the generated answer is correct. The probability would simply be what fraction of all generated answers was the one selected. (Each generated answer would be generated with a different seed and the question would be of single choice kind.) The two issues I found were 1) the cost, 2) on some problems, LLMs can be wrong more often than they are not.

Hopefully, as inference gets cheaper and of higher quality, someone will come up with a more feasible solution.

xcv1234• 2 years ago

I'm not sure people in these comments are reading this paper correctly. This seems to essentially disprove the whole idea of multi-agent setups like Chain-of-thought and LLM-Debate.

I'm not sure you have read the paper at all. Chain of thought prompting is not a multi-agent algorithm. The paper says that it enhances existing methods such as prompt engineering (chain of thought) and multi-agent debate. The sampling method presented in the paper is orthogonal to those methods.

0x008• 2 years ago

This is a very similar idea to ensemble models, which have been used for a long time in ML and proven to be very good. You average out the results of several predictors (or you let them vote and pick the most common prediction value), thereby reducing the noise in the prediction by choosing the common denominator of multiple predictions.

sroussey• 2 years ago

This is done in aerospace as well… however, even different teams clean room writing to the same spec have the tendency to make the same errors in their code, which ends up breaking the statistical model when this model was selected.

mirekrusin• 2 years ago

Good news is that you can use this setup for self supervised RL (artificial dreaming? increasing contrast?).

ben_w12• 2 years ago

Which makes sense to me. If an LLM is giving you a mixture of "hallucinations" and correct answers, the correct answers will similar and the hallucinations will hopefully be chaotic

I expect that to give you something close to the confidence of the underlying model to some specific claim, which is good, but I still expect legends (urban and cultural) to be high-ranked.

They'd be very human mistakes, but still mistakes.

I think the only way past that is to build a world model, look for contradictions, and then look for new evidence to resolve those contradictions.

dmarchand90• 2 years ago

Be interesting to plug this into a bayesian optimization like framework: find out regions of language space where the models maximally disagree and then target those areas for extra training

Paradigma11• 2 years ago

My impression from github copilot is that hallucinations are the result of certain true facts having a low likelihood and copilot giving you the most likely answer anyway.

Typically I have a certain library that does things in a very unorthodox and undocumented way and when I ask copilot for an example it gives me wonderful, totally understandable code of made up functions that I wouldnt need in the first place if the library worked that way.

I dont think that running that query multiple times would help.

rasz• 2 years ago

Which makes sense to me. If an LLM is giving you a mixture of "hallucinations" and correct answers, the correct answers will similar and the hallucinations will hopefully be chaotic

Not my experience. I had multiple LLMs hallucinate hard when asked same question multiple times. The only way to break the cycle is to follow everything with questions demanding clarifications. "are you sure?" "this is wrong, correct the answer".

auraai• 2 years ago

I don't think this type of method can scale indefinitely, it's essentially just "better" sampling within dense areas of knowledge space. It cannot help with better exploration outside these dense areas, because these explorations won't have a consensus among agents almost by definition.

kshitij_libra• 2 years ago

This trend really needs to die. If you can’t come up with an original paper name , maybe the contents aren’t that original either

parentheses• 2 years ago

I wonder if this performs even better when this is done tokenwise in the inner loop of the LLM

atum47• 2 years ago

If we sum all those "x is all you need" we're going to realize that we need a lot of things

trash_cat3• 2 years ago

Is this not an incredibly expensive/unsustainable method? I agree with the sentiment that MoE is the way to go as the newer models will probably see diminishing returns. But the compute for a single prompt will suddenly increase 7-15 fold?

bigEnotation2• 2 years ago

Yeah, see GPT 3.5 vs GPT 4 pricing.

bearjaws2• 2 years ago

"all you need is a six figure OpenAI bill"

Garlef• 2 years ago

if it replaces 2-3 experts that might be a good deal

ukuina2• 2 years ago

If GPT4 is 20x the price of GPT3.5, but it only takes 10x GPT3.5 runs to get similar quality of response (and likely faster), you'll still come out ahead.

dimask• 2 years ago

I doubt that 10xGPT3.5 > GPT4. There are a lot of tasks that GPT4 can do and GPT3.5 just cannot. Also, in such cases I find that GPT3.5's hallucinations are quite consistent, so such a method is probably not gonna help.

imtringued• 2 years ago

So what? It's not like GPUs are compute starved.

ZachSaucier• 2 years ago

Also the amount of non-renewable resources used and emissions goes up 7-15x

bigEnotation2• 2 years ago

I thought this is what GPT 4 was, it uses a boosting algorithm over GPT 3.5?

bearjaws2• 2 years ago

I'm sure all the AI companies love the idea of running the same prompt 8 times...

datascienced• 2 years ago

NVDA does for sure

sheepscreek• 2 years ago

Having given this problem a great deal of thought, I have developed a strong intuition around this. I believe not only is AGI feasible, it is already doable.

For example, several hundred GPT-4 based agents specializing in different skill sets should be able to collaboratively solve many problems. Their ability to work on so many facets of the same problem will make them very effective against multidisciplinary problems.

What’s the catch? Well, the back and forth has to play out in a serial order, so it cannot be parallelized. At today’s abysmal inference speeds, it may take this AGI many times longer than a trained human. Now imagine the effectiveness of this method when we can speed up inference to several hundred times a minute. Now AGI suddenly becomes way more efficient than a human.

29athrowaway• 2 years ago

There's a paper about that

https://www.microsoft.com/en-us/research/publication/sparks-...

intended2• 2 years ago

Definition of AGI at play here?

etamponi• 2 years ago

Isn't this the same as lowering the temperature of the LLM?

whiteandnerdy• 2 years ago

I remember hearing that Beam Search doesn't work well for LLMs, because it leads to repetitive, generic output.

The majority vote sampling technique in this paper sounds like it'd give similar output to Beam Search, because it's sampling sequences of tokens from a joint distribution. So why doesn't it give repetitive output like Beam Search does? What am I missing?

l_l_m_5_0• 2 years ago

Averaging LLM outputs will ensure the final output will contain a lot of words with no substance. However, it’s essential to recognize that averaging bad data doesn’t always lead to better results. Garbage in, garbage out — averaging cannot magically transform flawed inputs into accurate outputs.

yantrams• 2 years ago

This is my go to method for pretty much every hard problem that I'm forced to solve where I don't have the domain expertise / interest / time. The trick lies in coming up with a clever similarity metric that incorporates penalties etc. You can even go a level deeper and use multiple similarity algorithms and then poll on top of them. Here's a taxonomy extractor for text that I made using similar principles that is surprisingly as good as anything else that I've seen - https://dash.scooptent.com/text

dindobre• 2 years ago

Smells similar to Sutton's bitter lesson!

krystofee• 2 years ago

this study's got me worried. it goes against the hope that AGI won't just sneak onto the internet and do its own thing. with more work on making LLMs bigger and showing they can do more, the thought that this could actually happen in the future gets real scary, especially since i can run smaller versions on my own pc now.

zone411• 2 years ago

This is quite interesting because I've specifically tried this kind of basic ensembling for my NYT Connections benchmark and it didn't work. This is something everybody would try first before more complicated multi-step prompting, and yet since ChatGPT 3.5 I'm not aware of any papers showing that it works. It will be interesting to reproduce this result and learn more about how they set it up to make it work.

msoad• 2 years ago

What's interesting is that each run of the model tends to converge to a different "local maximum" in the solution space, and some of these local maxima correspond to better performance than others. By running the model multiple times, we increase the chances of finding a higher-quality local maximum or even the absolute best solution.

This got me thinking: why is this ensembling step implemented as a higher-level abstraction on top of the base LLM, rather than being built directly into the neural network architecture and training process itself?

sp3322• 2 years ago

Well you’re right that LLM tooling is totally inadequate. At least we already have beam search. But the more boring answer (and why beam search is also uncommon) is that running the query multiple times is more expense.

Takennickname• 2 years ago

The arrogance to name your paper that.

bwnjnoei• 2 years ago

Model ensemble is a classic method. Deep learning is always rediscovering and reinventing the classics. I believe that many people have used this method before, but they just haven't published a paper on it.

edshiro• 2 years ago

I've saved the paper to read it later.

The premise of this work seems very interesting... But I wonder how practical it is from both a cost and time perspective. I am toying around with an AI Agents library and one of the annoying UX things I notice is the time it takes to get my answers, because each call to an agent (either GPT-4 or Claude 3) is kinda slow.

Besides the time, it feels quite wasteful token wise.

I'm skeptical this approach will be adopted by many in the AI Agent space, but of course I could be very wrong.

temporarely2• 2 years ago

Just reading the (current top) few comments and whimsically wondered at the super business model of companies offering LLM services: a car service that won't get you from point A to B unless you hail it n times. A detergent that must be applied n times before cloths come out ("probably") clean.

If a company is offering "Artificial intelligence" at a price, then isn't reasonable that you only pay for correct answers? If the company is offering car service, shouldn't you only pay if they take you to your destination?

CSSer• 2 years ago

Agreed, and if it fails often enough, isn’t the bar at which a human or general-purpose, traditionally structured automation is going to be superior pretty low? This is how I think this bubble will pop. No doubt, LLMs are a breakthrough tool, but I’m sincerely skeptical of all but the most granular applications.

Perhaps the moral is that diffusing LLM agent accountability has the same failure model as the pre-existing human one.

nequo• 2 years ago

Counterpoint: National Weather Service forecasts don’t always turn out to be correct but we don’t only pay the NWS on days when they are correct.

samus9• 2 years ago

Companies usually offer a service or a product. If the company doesn't deliver what was agreed upon, then the customer can demand correction. If a taxi driver takes a needlessly convoluted route, charges too much, or doesn't bring you to the destination, you can complain to the taxi company. If the laundry didn't work, you insist on doing it again.

However, many activities are inherently fraught with risk or uncertain results since there are always things outside of anyone's control. A lawyer can't promise you prevail in a court case, but they have to advocate your case to the best of their abilities. A doctor won't guarantee that you become healthy again. No taxi driver will guarantee you that you will reach the destination in time, but they will bring you there. Atlassian won't guarantee you will meet a release deadline if you use their managed JIRA instance, but they will do their best to prevent data loss. And a company that basically sells access to a chatbot won't guarantee that it gives you correct results. Maybe availability guarantees.

temporarely2• 2 years ago

Reading the paper the thought occurs that IF the likelihood of correct response increases by the number of agents employed AND this involves application of a function (whatever) to select the 'best' from the possible answers, doesn't that imply that LLM has insufficient dimensions?

In other words, I am wondering if LLM hallucinations [sic] are in fact symptomatic of 'conflation' which could itself be the result of insufficient dimensions.

Thoughts?

adamlp• 2 years ago

https://anonymous.4open.science/api/repo/more_agent_is_all_y...

How can you have majority voting based on text produced by LLMs, won’t every answer string be essentially distinct with extremely high probability if it’s over a couple dozens bits of output (which any useful LLM output would be, right?)?

mateiz• 2 years ago

This is a cool paper showing there is value in using an LLM multiple times, but in recent research we showed that with majority voting, quality can decrease past some point as you make more calls. Check out https://arxiv.org/pdf/2403.02419.pdf. It raises the natural question of how to design the best inference algorithm given an LLM you can call multiple times.

More Agents Is All You Need: LLMs performance scales with the number of agents

Related Stories