Llama 3 Plus Groq CHANGES EVERYTHING!
ฝัง
- เผยแพร่เมื่อ 21 เม.ย. 2024
- **If you are looking to purchase a new Tesla Car, Solar roof, Solar tiles or PowerWall, just click this link to get up to $500 off! www.tesla.com/referral/john11286. Thank you!
Join this channel to get access to perks:
/ @drknowitallknows
Check out Matthew Berman's recent video: • LLaMA 3 “Hyper Speed” ...
**To become part of our Patreon team, help support the channel, and get awesome perks, check out our Patreon site here: / drknowitallknows . Thanks for your support!
Get The Elon Musk Mission (I've got two chapters in it) here:
Paperback: amzn.to/3TQXV9g
Kindle: amzn.to/3U7f7Hr!
**Want some awesome Dr. Know-it-all merch, including the YEAR OF EMBODIED AI Shirt? Check out our awesome Merch store: drknowitall.itemorder.com/sale
For a limited time, use the code "Knows2021" to get 20% off your entire order!
**Check out Artimatic: www.artimatic.io
**You can help support this channel with one click! We have an Amazon Affiliate link in several countries. If you click the link for your country, anything you buy from Amazon in the next several hours gives us a small commission, and costs you nothing. Thank you!
* USA: amzn.to/39n5mPH
* Germany: amzn.to/2XbdxJi
* United Kingdom: amzn.to/3hGlzTR
* France: amzn.to/2KRAwXh
* Spain: amzn.to/3hJYYFV
**What do we use to shoot our videos?
-Sony alpha a7 III: amzn.to/3czV2XJ
--and lens: amzn.to/3aujOqE
-Feelworld portable field monitor: amzn.to/38yf2ah
-Neewer compact desk tripod: amzn.to/3l8yrUk
-Glidegear teleprompter: amzn.to/3rJeFkP
-Neewer dimmable LED lights: amzn.to/3qAg3oF
-Rode Wireless Go II Lavalier microphones: amzn.to/3eC9jUZ
-Rode NT USB+ Studio Microphone: amzn.to/3U65Q3w
-Focusrite Scarlette 2i2 audio interface: amzn.to/3l8vqDu
-Studio soundproofing tiles: amzn.to/3rFUtQU
-Sony MDR-7506 Professional Headphones: amzn.to/2OoDdBd
-Apple M1 Max Studio: amzn.to/3GfxPYY
-Apple M1 MacBook Pro: amzn.to/3wPYV1D
-Docking Station for MacBook: amzn.to/3yIhc1S
-Philips Brilliance 4K Docking Monitor: amzn.to/3xwSKAb
-Sabrent 8TB SSD drive: amzn.to/3rhSxQM
-DJI Mavic Mini Drone: amzn.to/2OnHCEw
-GoPro Hero 9 Black action camera: amzn.to/3vgVMrH
-GoPro Max 360 camera: amzn.to/3nORGYk
-Tesla phone mount: amzn.to/3U92fl9
-Suction car mount for camera: amzn.to/3tcUfRK
-Extender Rod for car mount camera: amzn.to/3wHQXsw
**Here are a few products we've found really fun and/or useful:
-NeoCharge Dryer/EV charger splitter: amzn.to/39UcKWx
-Lift pucks for your Tesla: amzn.to/3vJF3iB
-Emergency tire fill and repair kit: amzn.to/3vMkL8d
-CO2 Monitor: amzn.to/3PsQRh2
-Camping mattress for your Tesla model S/3/X/Y: amzn.to/3m7ffef
**Music by Zenlee. Check out his amazing music on instagram -@zenlee_music
or TH-cam - / @zenlee_music
Tesla Stock: TSLA
**EVANNEX
Check out the Evannex web site: evannex.com/
If you use my discount code, KnowsEVs, you get $10 off any order over $100!
**For business inquiries, please email me here: DrKnowItAllKnows@gmail.com
Twitter: / drknowitall16
Also on Twitter: @Tesla_UnPR: / tesla_un
Instagram: @drknowitallknows
**Want some outdoorsy videos? Check out Whole Nuts and Donuts: / @wholenutsanddonuts5741 - บันเทิง
Give it concrete steps. For example say this is a two step process. Step 1: formulate 10 solutions. Step 2: evaluate the solutions from step 1 and select the best one. It helps to understand that you are writing a program. The models don’t yet seem to deduce concrete steps sometimes.
I discovered groq recently (April 20?) and there you are, already reporting on it!! Amazing, Mr know it all. You do know it all!
Also, interesting is that Groq, seemingly, can help reduce the black box phenomenon by explaining its decision making process and be able to evaluate its own reasoning.
It perhaps says that's the process without knowing the process. It may be saying what others think the process should be.
You earned a subscriber from Kenya. Keep going. I guess we are age mates. Very solid content.
Whatever i want, whatever i like, whatever i need.
I've not done the video yet, but you may want to talk about the differences between grok and groq! I suspect many may mix them up!
Yeah. What's what here?
Also lets not forget about Grok the common Regex style pattern parsing often used to parse logs. Not to be confused with Grog the alcoholic drink
Because nobody has ever purchased a "lite" salad dressing.
Can you include the 10 answers literally in the next question?
Fascinating work! thanks for the vid.
Anyone else annoyed with the Groq / Grok name conflict?
At least Grok retains the original spelling given by Heinlein.
Nah . . .
It confuses me every time
@@taiwanjohnThey sound the same, but XAi can't claim a trademark because "grok" is in the dictionary and (I think) "groq" isn"t.
I would have been more annoyed if they had tried to hijack grok
cup and marble are not defined sizes. Try with description of marble size relative to cup's interior diameter (or actual sizes). Marble is a sphere, but it is also a material.
What would be nice is if would 1. ask for clarification or 2. state assumptions in preamble to answer.
Ask it to suggest improvements of your pre-prompt
It occurs to me that the "self reflection" approach seems a lot like an adversarial network. Perhaps additional training with a gan would improve the model overall?
Good God Doc, were you actually typing your question that quickly?
When I tried to match your typing speed...I didn't recognise what language I had written it in!
I find it gets better math responses if I have it explain how to solve the math question for a student.
Perplexity's answer to "Who makes h100 chips" is correct.
In your question, can you ask it to take more time to determine a better answer? I believe some other models will do this.
I think Groq with agentic workflows, is a good combination. (Andrew Ng, who is more conservative at the AI end, also thinks agentic worflows will be the next big thing.). Even if it succeeds, I'm always leery of LLM's doing "reasoning". It's a bit dicey. It's not their forte as natural language models, associative auto-complete engines. A good evaluator in the workflow would help. And as stated in the video, based solely on language training, it doesn't have the embodiment, implicit knowledge, to always answer the question correctly. One can help fill in the gaps, with more in-context information.
Using this as a system prompt, not a pre-prompt, allowed a single shot correct response to the algebra question. You could probably just use this as a pre-prompt, but I don't like typing it over and over so I just set it as the system prompt. :-)
"You will engage in silent reflection before providing a final response to all prompts. This means you will silently consider at least 5 preliminary responses to each prompt, which preliminary responses may be similar or dissimilar in whole or part from one another. You will then silently reflect upon each preliminary response in consideration of the other preliminary responses. You will then derive your final response incorporating the learnings. clarification, corrections and valuable insights gained by your silent reflection."
It also got the marble question right the first time. 🙂
Thank you for sharing this
Simple reasoning question no model gets right:
William has 2 brothers and 3 sisters. Which of the following could be a sibling of William?
Michael: has 3 brothers and 2 sisters
Sarah: has 2 brothers and 3 sisters
Jimmy: has 1 brother and 3 sisters
Emma: has 3 brothers and 2 sisters
Gender inference from names is not a thing these days.
Maybe I’m getting behind in keeping up with LLM‘s and how they work, but doctor know it all, please explain if you can how a large model can reason at all, not just arrive at correct answers. How does predicting the next word in a sentence produce reasoning? Or does it have to do with transformers, which I’m kind of fuzzy on. Maybe you can either cover this in a future video, or reference a previous video where you have already covered it. Thanks so much.
Transformers work almost exactly the same as the human brain. That's how.
Essentially you have tokens(neurons) that pass through to different sections that generate different outcomes. When your brain predicts the best outcome that outcome is then executed.
Groq need Grok as model in the list.
Any chance you can share your pre-prompt?
And if you think about how humans solve these problems, that involves a lot more runs with different prompts, and many differently fine-tuned models too.
Interesting!
Agents+relfexion+Groq = 🎉
This works better:
"You will engage in reflection before providing a final response to all prompts. This means you will consider at least 5 fully complete preliminary responses to each prompt, which preliminary responses may be similar or dissimilar in whole or part from one another. You will then reflect upon each preliminary response in consideration of the other preliminary responses. You will then derive your final response incorporating the learnings. clarification, corrections and valuable insights gained by your silent reflection."
It gets correct one-shot responses with Llama3 70b. I also gave it the below question and it was one-shot correct. It was interesting that some the preliminary responses were incorrect ones I have seen before, and on reflection it figured out the correct answer among the preliminaries:
"I dried a shirt outside in the sun and it took 1 hour. How long will it take me to dry 5 of the exact same shirts outside in the sun under the exact same conditions?"
Just tried this on Llama3 8b using Ollama on my Laptop with an RTX4080. Holy shmokes it was fast, but it got it wrong.
">>> You will engage in reflection before providing a final response to all prompts. This means you
... will consider at least 5 fully complete preliminary responses to each prompt, which preliminary
... responses may be similar or dissimilar in whole or part from one another. You will then reflec
... t upon each preliminary response in consideration of the other preliminary responses. You will
... then derive your final response incorporating the learnings. clarification, corrections and val
... uable insights gained by your silent reflection. Assume the laws of physics on Earth. A small
... marble is put into a normal cup and the cup is placed upside down on a table. Someone then ta
... kes the cup without changing it's orientation and puts is inside the microwave. Where is the m
... arble now? Explain your reasoning step by step.
I'll follow the process you've outlined.
**Preliminary Responses**
1. **Initial Response**: The marble remains at the bottom of the cup inside the microwave.
* Reasoning: Gravity keeps the marble stuck to the bottom of the cup, and the rotation of t
the microwave doesn't affect its position.
2. **Modified Initial Response**: The marble slides up the inside of the cup due to
centrifugal force.
* Reasoning: The rotation of the microwave creates a force that pushes the marble upwards,
but gravity still holds it in place at the bottom of the cup.
3. **Alternative Scenario**: The marble rolls out of the cup and onto the turntable or floor
of the microwave.
* Reasoning: As the cup rotates, the marble could be dislodged from its position and fall to
the floor due to the centrifugal force.
4. **Cup Orientation Matters**: The marble stays at the bottom of the cup because the cup is
still upside down.
* Reasoning: Since the cup remains in an inverted position, gravity continues to hold the m
marble firmly at the bottom.
5. **Microwave Rotation Dominates**: The marble is flung out of the cup and scattered around
the interior of the microwave.
* Reasoning: The rapid rotation of the microwave overcomes any gravitational force, sending
the marble flying out of the cup.
**Reflection**
As I reflect on these preliminary responses, several insights emerge:
* Gravity plays a significant role in determining the marble's position, but it can be
overcome by centrifugal force and other forces.
* The orientation of the cup affects the marble's movement and final position.
* The rotation of the microwave can have a profound impact on the marble's trajectory.
**Final Response**
Considering these insights, I derive my final response:
The marble rolls out of the cup and onto the turntable or floor of the microwave. As the cup
rotates in the microwave, the centrifugal force dislodges the marble from its position at the
bottom of the cup. Gravity initially tries to hold it in place, but the rapid rotation of the
microwave overcomes its influence, sending the marble flying out of the cup.
By engaging in reflection and considering multiple perspectives, I've gained a deeper
understanding of the problem and arrived at a more comprehensive answer."
Fascinating response nonetheless. Sitting here dumbfounded how in less than 10 minutes I was able to install and run an LLM locally so quickly. Never tried it before. Thanks for piquing my interest to try this.
This all new to me. Think it is time to learn. I Googled Grog but only crap answers. How, where do you get Grog? Thank You.
Grog😂
Careful groq spat out some wrong answers for me today. I asked it to correct itself after giving it instructions, which it promised to do, then failed again, I then told it had failed and why, then they suggested I do an internet search.
😂
AI really lacks the ability to operate data. Sure, it can predict it, but it can't do much else with it. Meanwhile humans have the ability to imagine (as long as you don't have an aphantasia) and just go into your memories or feelings, navigate the brainmap, focus on your senses and all. So yeah, it really needs some sort of collection of navigatable and reconstructable structures, closely tied into the very neural structure of an AI, which it could use to recall whatever vague semantics it had generated during inference. Final text alone can't store those.
Did you state in your question that the cup did not have a cap or lid?
Very interesting
Anyone know if P.O.E. uses, or plans to use, Groq?
Put parenthesis around the first two terms
So shortly put talking to this AI is currently kinda of like talking to a schizophrenic Albert Einstein that produce a lot of hallucinations, but by prompting it to give 10 answers then force it to reevaluate and filter out the hallucinations it has a high chance of ending up with the correct answer. 😎👍❤️ as a software engineer with advanced reverse-engineering as my focus, I am very impressed by your approach.
Can Tesla use groq?
Can’t wait until we get a Cortana-level AI then it’ll respond with some banter and sass 😂
Why on earth wouldn't you list your prompts in the description or in the first comment? Especially when you're asking for suggestions for improvement? Since you didn't, I can't easily try and see what would help. If I could, I would have tried asking to both show work AND verify the answer.
👍
I'm not sure that methodology is valid with the current models. If you ask it to generate 10 answers and pick the best one with one prompt, that may be different than first prompting for ten answers and then making a new prompt to look at its last answer and pick the best one. When it does it all in one prompt, it probably can't "back up" to look at the list, so it's "best answer" is generated more from your question prompt rather than the list. When you do a second prompt, then the list becomes part of the new prompt input, and it can reasonably consider the 10 answers.
It's literally just a feedback loop with a human doing the feedback. It's something the model already does it just doesn't have this prompt built in. Future fine tunes will I'm sure.
2/(a-1)
Groq is dumb. Ask it: "List some ten letter words." For instance.
I asked it to name three words with ten letters and it worked.
@@m.3257 It doesn't work.
prompt: Please list some ten letter words.
Here are ten 10-letter words:
Recommits
Encounter
Conversed
Carriages
Gathered
Informing
Occurred
Shattered
Whispered
Calculus
Here's some more
prompt: Find a 10 letter word in which each letter appears only one time.
That's a great challenge!
After searching, I found a 10-letter word that meets the condition: "uncopyright". Each letter in this word appears only once.
Here it is: u-n-c-o-p-y-r-i-g-h-t
Damn, Doc, Let's talk about the $100/share price drop since January, and another $4k drop in the purchase price of FSD. $142 is looking to look high after 4/23
You don't seem to understand that inference is forward only, an LLM can't go back and review its answers before doing a final answer in 0 shots. You have to do multi-shot feedback for the LLM to read its own thoughts. Your prompts simply cannot work, you're asking for something an auto-regressive forward only LLM can't do. The second shot is effectively a RAG query with the previous answers being fed back in as new data where it then sees and chooses from the additional data. LLM's cannot ponder their own output in 0 shots.