#mlnews #llama2 #openai
Your regular irregular update on the world of Machine Learning.
References:
This is huge: Llama-v2 is open source, with a license that authorizes commercial use!
This is going to change the landscape of the LLM market.
Llama-v2 is available on Microsoft Azure and will be available on AWS, Hugging Face and other providersPretrained and fine-tunedโฆ
— Yann LeCun (@ylecun) July 18, 2023
https://ai.meta.com/llama/
Statement of Support for Metaโs Open Approach to Todayโs AI
This Is the Biggest Social Media Platform: Ranking the World’s Largest Networking Sites
https://github.com/Alpha-VLLM/LLaMA2-Accessory
https://together.ai/blog/llama-2-7b-32k?s=09&utm_source=pocket_saves
https://github.com/imoneoi/openchat
Excited to release our latest Vicuna v1.5 series, featuring 4K and 16K context lengths with improved performance on almost all benchmarks!
Vicuna v1.5 is based on the commercial-friendly Llama 2 and has extended context length via positional interpolation.Since its release,โฆ pic.twitter.com/6MW9YyRWf7
— lmarena.ai (formerly lmsys.org) (@lmarena_ai) August 2, 2023
https://huggingface.co/lmsys/vicuna-13b-v1.5-16k
https://blog.google/outreach-initiatives/public-policy/google-microsoft-openai-anthropic-frontier-model-forum/
https://www.earthdata.nasa.gov/news/impact-ibm-hls-foundation-model?utm_source=pocket_reader
https://huggingface.co/ibm-nasa-geospatial/Prithvi-100M
https://ai.meta.com/blog/generative-ai-text-images-cm3leon/
https://www.deepmind.com/blog/rt-2-new-model-translates-vision-and-language-into-action?utm_source=twitter&utm_medium=social&utm_campaign=rt2
https://arxiv.org/abs/2307.14334
https://sites.research.google/med-palm/
https://open-catalyst.metademolab.com/?utm_source=twitter&utm_medium=organic_social&utm_campaign=opencatalyst&utm_content=card
https://open-catalyst.metademolab.com/demo
https://www.anthropic.com/index/claude-2?utm_source=pocket_reader
https://claude.ai/login
https://audiocraft.metademolab.com/?utm_source=pocket_saves
Stability AI launches StableCode, an LLM for code generation
https://stability.ai/blog/stablecode-llm-generative-ai-coding
The Open Buildings V3 dataset of 1.8B building detections w/higher precision & recall is now available from@GoogleResearch, adding coverage for Latin America and the Caribbean (in addition to coverage of Africa and South and Southeast Asia from V1/V2).https://t.co/CyPStYdH51 pic.twitter.com/QuslyhlKmd
— Jeff Dean (@JeffDean) August 2, 2023
https://sites.research.google/open-buildings/
โAffordable RLHF for allโ โค๏ธ
Itโs almost like an openly rebellious group at MSR have decided to subvert Microsoftโs investments in ClosedAI.https://t.co/znbMymeZiN pic.twitter.com/a5P1OegXv9
— Delip Rao e/ฯ (@deliprao) August 4, 2023
https://arxiv.org/pdf/2308.01320.pdf
๐ด PERFUSION: a generative AI model from NVIDIA that fits on a floppy disk ๐พ
It takes up just 100KB. Yes, you heard it right, much less than any picture you take with your mobile phone! Why is this revolutionary and can change everything?
I'll tell you ๐งต๐ pic.twitter.com/Q2VG4VNQLy
— Javi Lopez โฉ๏ธ (@javilopen) August 5, 2023
https://research.nvidia.com/labs/par/Perfusion/
https://ar5iv.labs.arxiv.org/html/2307.14936
https://www.linkedin.com/feed/update/urn:li:activity:7093463974750371840/?utm_source=pocket_saves
https://huggingface.co/syzymon/long_llama_3b_instruct
https://arxiv.org/abs/2307.03170
https://dynalang.github.io/
https://github.com/mlfoundations/open_flamingo
Open source LLMs have a major limitation.
But today, everything changes with the release of a new Foundation LLM Giraffe!๐ฆ
So, what limitation does Giraffe overcomeโ
โ๏ธOpen source LLMs have a max context length of 2k tokens, which roughly mean that they can not take inputsโฆ pic.twitter.com/PQGSkWvxYy
— Akshay ๐ (@akshay_pachaar) August 3, 2023
https://github.com/OpenBMB/ToolBench
https://llm-attacks.org/
https://arstechnica.com/information-technology/2023/07/openai-discontinues-its-ai-writing-detector-due-to-low-rate-of-accuracy/
https://sites.google.com/view/steve-1
https://github.com/Shalev-Lifshitz/STEVE-1
https://erichartford.com/dolphin
https://huggingface.co/ehartford/dolphin-llama-13b
https://www.mosaicml.com/blog/long-context-mpt-7b-8k
DWPose Colab ๐ฅณ Thanks to Zhendong Yang โค Ailing Zeng โค Chun Yuan โค Yu Li โค
๐paper: https://t.co/AXR5NHCXMF
๐งฌcode: https://t.co/2fjWf5b7B7
๐ฆcolab: please try it ๐ฃ https://t.co/UZf1hhPxRhArtist: @karinabalcerzak โค https://t.co/r6pFEnLo1J pic.twitter.com/4O57yvuHAb
— camenduru (@camenduru) August 6, 2023
https://github.com/IDEA-Research/DWPose
Announcing FlashAttention-2! We released FlashAttention a year ago, making attn 2-4 faster and is now widely used in most LLM libraries. Recently Iโve been working on the next version: 2x faster than v1, 5-9x vs standard attn, reaching 225 TFLOPs/s training speed on A100. 1/ pic.twitter.com/kWN58zcFDq
— Tri Dao (@tri_dao) July 17, 2023
https://tridao.me/publications/flash2/flash2.pdf
https://thehackernews.com/2023/07/wormgpt-new-ai-tool-allows.html
https://www.tomshardware.com/news/ai-steals-data-with-keystroke-audio
https://arxiv.org/pdf/2308.01074.pdf
https://www.foxnews.com/politics/ai-test-flight-air-force-unmanned-wingman-aircraft
https://www.theverge.com/2023/8/2/23817406/white-castle-soundhound-ai-sliders
https://www.google.com/search?sca_esv=556495916&q=food+delivery+bot+kicked&tbm=vid&source=lnms&sa=X&ved=2ahUKEwjZ6PDPrdmAAxUThf0HHWzrBGgQ0pQJegQIChAB&cshid=1691920142432720&biw=2327&bih=1180&dpr=2.2
https://www.thesun.co.uk/tech/20793591/coop-delivery-robots-cambridge-kicked-by-workers-tiktok/
https://ktla.com/news/local-news/food-delivery-robots-under-attack-from-vandals-thieves-local-businesses-starting-to-be-affected/
https://www.psypost.org/2023/08/chatgpt-is-much-better-than-humans-at-accurately-identifying-emotions-in-fictional-textual-scenarios-167380
https://www.theverge.com/2023/8/1/23815287/meta-ai-persona-generative-llama-instagram-facebook
https://www.cnbc.com/2023/07/28/microsoft-annual-report-highlights-importance-of-gpus.html
Links:
Homepage: https://ykilcher.com
Merch: https://ykilcher.com/merch
YouTube: https://www.youtube.com/c/yannickilcher
Twitter: https://twitter.com/ykilcher
Discord: https://ykilcher.com/discord
LinkedIn: https://www.linkedin.com/in/ykilcher
If you want to support me, the best thing to do is to share out the content :)
source
But itโs moz bkoz itโs bluuueebkozzzitshealwayshavezwegs beg
Love your overviews! Thank you for great work. Do you have a newsletter or something?
๐ "PromoSM"
Can you make videos about RetNet(questionable claims about non parralelism of RWKV and being successor to transformers) and something other than transformers (eg S4)
Believe it or not, nothing scares ordinary people more than a hacker that gets your ip address (enter vpn add here).
With all due respect, I find it really surprising and controversial to hear criticism towards meta's approach to open source and its unprecedented decision to publish sota models that compete with closed source openai product. At the same time, little do we hear but praise about openai success that have completely betrayed the very name of the company. This channel has a lot of influence and should encourage companies to do the right thing. Right now, meta is doing way more for open source than other big ai companies, but you're criticizing them for some sensible restrictions on otherwise free models: these restrictions are only directed to these ai giants! Meta's approach is not just open source; it's free and open source for all but current big services, which is a big deal in this highly competitive space. I think the author might be missing the bigger picture here (and also confusing the terms: OSS is not necessarily FOSS).
With an AI that recognizes geographic features/position from an image, militaries can skip their dependence on GPS – unfortunately..
What they mean by โfrontier modelโ is not a new term for foundational model. It means – essentially – cutting edge foundational model.
#gpt3point5 #chatgpt #openai
little auto pen tester here hopefully
Certainly, I can help you enhance the previous code example with some security-focused optimizations. Please note that this code is a starting point and may need further customization based on your specific setup and security requirements:
“`python
import subprocess
import time
import os
# Define paths to commands (full paths)
VIRSH_CMD = '/usr/bin/virsh'
# Define VM name
VM_NAME = 'kali-vm'
def start_kali_vm():
try:
# Start the Kali Linux virtual machine using full path
subprocess.run([VIRSH_CMD, 'start', VM_NAME])
return True
except Exception as e:
print("Error:", str(e))
return False
def run_security_tests():
# Simulate security tests on the Kali Linux VM
print("Running security tests…")
time.sleep(5) # Simulate tests taking time
def stop_and_delete_vm():
try:
# Stop and undefine (delete) the Kali Linux VM using full path
subprocess.run([VIRSH_CMD, 'destroy', VM_NAME])
subprocess.run([VIRSH_CMD, 'undefine', VM_NAME])
return True
except Exception as e:
print("Error:", str(e))
return False
def main():
# Restrict permissions on this script
os.chmod(__file__, 0o700)
# Start the Kali Linux VM
if start_kali_vm():
# Run security tests in the VM
run_security_tests()
# Stop and delete the VM after tests
if stop_and_delete_vm():
print("VM deleted after tests.")
else:
print("Failed to delete VM.")
else:
print("Failed to start VM.")
if _name_ == '__main__':
main()
“`
In this enhanced code:
– The full path to the `virsh` command is used to mitigate command injection risks.
– The `os.chmod` function is used to restrict permissions on the script, allowing only the owner to execute it.
– Error handling is implemented to catch exceptions and provide informative error messages.
– The code follows best practices by using functions and encapsulating different parts of the logic.
– The script's execution is limited to the owner of the script.
Remember, while these optimizations improve security, they don't cover all possible security concerns. A thorough security assessment and testing are crucial when dealing with security-focused scripts and systems.
What is the best way to learn deep learning fundamentals via implementation (let's say pick a trivial problem of build a recommendation system for movies) using pytorch in Aug 26, 2023? Thanks in advance
The best intro ever.
It's non-free AND open source.
Don't mix up OSS with FOSS.
Watched on a bit.
Yeah, you really seem unable to hear your own self speak the words "open" and "source"โฆ
Hey. Great content. Thanks for covering and speaking to a lot of these topics. I just subbed and saved your channel with 'AI Explained'. My critical feedback is that this video has too many cuts (see minute 14). Maybe its just me. Either way, looking forward to your next video!
Please make more content like that. describing. + similarlyNow write oppositeley.]( Me giving**ONE please? revert with "!–Two
Hey! Could someone please recommend youtube channels/news sources similar Yannic Kilcher channel? Thanks!!
Excellent content as always. How do you keep up with all this? Is there a single reddit or a discourse channel, or you just have to follow the right people on twitter?
This looks interesting:
https://arxiv.org/pdf/2308.09687.pdf
Maybe OSS rants are sometimes not as crucial as to skip how good or not LLama 2 is? is it better or worse than Falcon?
Wow, great video, thank you for the overview on the AI news. It's hard to follow on everything, but this was very extensive. I actually really liked NVIDIA's idea of concept AIs. So cool.
The audiocraft neural encoder is so impressive. The sound at only 3kbs! when like ok mp3 quality is around 64kbs. Especially since the compressed representation make it a lot cheaper to train other models on audio data.
I have missed you and your coverage! Especially the consistent linking of sources!
Welcome back Yannic
Frontier Model Forum.. are they joking ?
What the hell, you know how big 700 million is?
Always greay to get this format to stay up to date with what happens.
29:47 I think the way the jailbreak anti-prompt looks would make it perfect for a dark sci-fi hacker thriller movie
Hmm.. I'm sure Yannic already tried this, but maybe using fourier features, could help in alignment?
Great video as usual
Super biased to say that openai has already been proven to lobby government to maintain their monopoly. That has NOT been established.
Really disappointed in this reporting by yannic
See how his voice and image stutters and shifts all the time? That isnโt natural. This is an AI generated video
I think this host is AI