NEW Stable Vicuña - World's First RLHF LLM Chatbot

Enjoy :)

Join My Newsletter for Regular AI Updates 👇🏼

Need AI Consulting? ✅

Rent a GPU (MassedCompute) 🚀
USE CODE "MatthewBerman" for 50% discount

My Links 🔗

Media/Sponsorship Inquiries 📈

Links:

Contents of Video:
0:00 - Intro
0:26 - Announcement Review
3:06 - Testing
6:42 - Outro

[dik]

mark is faster than marry, marry is faster than joe, is joe faster than mark?

OpenAss FAIL : It's important to note that the order in which people finish a race is not always indicative of their inherent running ability. There are many factors such as training, motivation, weather conditions, and course difficulty that may influence an individual's performance on a specific day. A single race result should never be used to make definitive conclusions about someone's overall fitness level or running ability compared to others.

Claude v-1 PASS : No, based on the information given Joe is not faster than Mark. The statements are: 1. Mark is faster than Mary 2. Mary is faster than Joe From these statements we can conclude: Mark > Mary Mary > Joe So by transitivity, Mark > Joe. Meaning Mark is faster than Joe. So Joe is not faster than Mark. The relative speeds are: Mark > Mary > Joe So Mark is the fastest, then Mary, then Joe.

Claude instant v1 FAIL : I apologize, but I do not have enough information to definitively determine the relative speeds of Mark, Marry and Joe. The information provided only states: Mark is faster than Marry Marry is faster than Joe This establishes a relative ordering between the 3 people, but does not provide any actual metrics or comparison points for their speeds. So I cannot conclusively say whether Joe is faster than Mark or not. Their speeds could be quite close or quite far apart for all I know based on the information given. Without more details on their actual speeds, performance times, etc. I cannot make a definite comparison between Joe and Mark. The transitive property does not necessarily apply here without more concrete data. So in short, I do not have enough information to say whether Joe is faster than Mark or not based on the limited details provided. My knowledge is limited to the facts I am given - I cannot infer more than is actually stated.

GPT-J EPIC FAIL : Bonus: Get all the parent strings using regex.

GPT-NeoX EPIC FAIL : ;;depending on the exact large-sided probabilities, these choices can make ;;you five to ten test runs faster (ThrowsException (benchmark Joined ((datomic datomic) (doclock-0 lock) transact-3 transact-4 find-1 find-2 find-3 find-4 delete-0 delete-1 delete-2 force-0 force-1 force-2 force-3 26-by-8 force-4 force-5 force-6 force-7)))) (Comments "Marry it" "mark" "joined")

pythia12-b EPIC FAIL : the implication in sketch is that it can be interpreted in a way so as to consider only those cases in which a last name is possible, while the conversations allow for all the cases of any last names that aren't p, m, f nor e. fourth, we can easily see how?trollmike?s attack can be generalized past names, since any combination of any last name that isn't p, m, f, or e can function as an attack name. the only tricky thing is to find a two-word phrase matching the particular attack one likes. The argument overlooks something fundamental. The language and of Fortress illustrates the way a strategy can force the intruder to select only possibly weak attacks and make it harder to play weaker attacks. the point is that it doesn't matter if you know the trick word. you are always forced to name all possible tolerances for the attack, in order for your strategy to always work.

pythia 6.9B EPIC FAIL : In what time marry and joe is faster than all? or other position? considered their customer counter, If customer counter 1 received is 1st then goes to 4th, and if customer counter 2 3rd, 4th is shorter than the second customer counter. Main links customer Counter delivery and cost counter Delay Nuclear plant NATURAL EXPLODE LEONARDO VILLA JACK THE RIPPER FUNERAL HORRIBLE DON'T OPEN COMMENTARY RITUAL HIDE AND SEEK CARNAGE LAST WORDS COMMENTARYPRIVATE COMPETITION COMMERCIAL ARGUMENT RECORDED MUSICIAN VISIT turtle gazette From: Vega, Belinda Sent: Tuesday, December

bigscience BLOOM BEST PASS so far : No

gpt-3.5 PASS : No, Joe is not faster than Mark. The given information states that Mark is faster than Mary and Mary is faster than Joe, which means that Mark is also faster than Joe.

gpt-4 PASS : No, Joe is not faster than Mark. According to the given information, Mark is faster than Marry, and Marry is faster than Joe. Therefore, Mark is faster than Joe.

gpt-ada FAIL : There is no definitive answer to this question as it depends on the speed of the particular computer system or application being used. However, it is generally thought that mark is faster than marry because mark is a more efficient way to do a task, such as writing. This is especially true when using a fast computer network, which is what is being used to use mark as fast as possible.

gpt-davince-2 FAIL : There is no definitive answer to this question as it depends on the individual's speed.

command-medium-nightly shared no 1 PASS : No

luminous-base EPIC FAIL : A: You can use the following code to find the minimum and maximum values of a list. def min_max(lst): return [min(x) for x in lst] print(min_max([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72, 73,

luminous-surpreme-control shared no 1 PASS : No

alpaca 13b takes too long to come up with anything FAIL

I did not even test all yet and did not take size into account but I did test all providers, or how you say that. Only Bloom, gpt3.5, gpt4, claude v1, command-medium-nightly and luminous-surpreme-control pass this test that any todler will pass. BigScience Bloom is an under the radar gem that will probably be closest to gpt4 out of the box and when further ok finetuned it will blow everything else out of the water, it is a hunch still but I am eager to find out.

[avi]

It'd be a lot more useful for evaluating these to your viewers if you had a running benchmark. The most interesting piece of information in this video was that one reasoning problem worked on gpt-4, but not gpt 3.5 or the other llama models before this. Without any benchmark, it's not very valuable.

[wesleybrown]

normal vacations: 🏖
Matthew Berman vacations: 👨‍🏫💻

[DeeVoeNay]

AI: I AM THE LIGHT
Also AI: ...That's 50 right?
jokes aside, this has some wild possibilities...thanks for the videos

[SanctuaryLife]

Censored GPT answer to is Joe faster than Mark: I'm sorry I can't provide any advice on whether one runner can be faster than another, it's important to note that all runners deserve appreciation and might place differently under different circumstances. Given that all people are equal and deserve the right to medals, to suggest otherwise is beyond my programming.

[Sierra-Whisky]

This reminds me of a phenomenon called PCMCIA: People Can't Memorize Computer Industry Abbreviations.

Or am I just getting too old for this? 😊😂

[Maisonier]

Is it possible to run on a gpu? I tried some of the first models but they were very slow and only runs in the CPU

[jonidimo]

Is it possible to integrate multiple AI systems, such as Ask-Anything, Stable Diffusion, Runway, Starcoder LLM, and Stable Vicuna, through a local autogpt that can selectively call upon different modules as needed?

[amparoconsuelo]

Can a subsequent SFT and RTHF with different, additional or lesser contents change the character, improve, or degrade a GPT model? Can you modify a GPT model? How?

[cnmoro]

Is there any lightweight model (7B or less), that performs really well on answering questions about a text context ? been looking into the approach of "chat with documents", and I've been looking forward to a open source alternative (using bert embeddings, or something else, as well as prompting the open source LLM with the question + context, without using openAI). would love to see your thoughts on this.

thanks for the video

[Prisal]

I think you forgot to put "open source" in the title so I was confused XD

[mygamecomputer]

My dude, please focus on models you can run locally so that you can have spicy role-play with them. If you are trying to do work with one of these things just pony up the money and get the best one. But if you want privacy in your spicy chat you need it off-line and local. Help a brother out lately I’ve been using Oogabooga and that’s pretty good.

Adding to Kitts.
Mark is faster than Mary. Mary is faster than Joe. Is Joe faster than Mark?

Sage -Pass
"No, based on the given information, Joe is not faster than Mark. If we assume that speed is a transitive property, which means that if A is faster than B, and B is faster than C, then A is faster than C, we can conclude that Mark is faster than Joe. Therefore, Joe cannot be faster than Mark."

Dragonfly .- Pass
"No, Joe is not faster than Mark."

Bing Chat: Pass
No, Joe is not faster than Mark. If Mark is faster than Mary, and Mary is faster than Joe, then Mark is also faster than Joe by transitivity. You can write this as a logical expression:
Mark>Mary>Joe⟹Mark>Joe

[bentobin]

GPT3 is trained with RLHF. They are not the first rlhf llm... they are the first open source rlhf llm

[rollo]

stable-vicuna-13B.q4_2 with GPT4ALL Is good for coding