This AI newsletter is all you need #24
Last Updated on December 22, 2022 by Editorial Team
This issue is brought to you thanks to Verta AI:
We’re happy to share The State of Machine Learning Operations research report by Verta AI Insights, which reveals a significant shift in how organizations are prioritizing their investments in AI/ML. Importantly, the ability to build complex models is no longer the competitive differentiator — organizations now have to effectively operationalize AI to stay competitive.
Read the report to learn about the trends and opportunities currently shaping the industry — and a movement towards an Operational AI mindset.
What happened this week in AI by Louis
Once again… OpenAI dominated the news section this week! Both with the amazing publication of Text-Davinci-003 and also, as you’ve certainly seen online, ChatGPT.
As we discussed with the community, Chat-GPT is pretty good from what we’ve seen so far. It may not be able to give you the right answer but it does seem to be able to understand what you are asking very well. Just look at this example shared by runoob#9765 on our Discord.
Here, we can clearly see that the model understands the question and even what’s implied (completing coding problems and being efficient enough to do a lot of them within a short timeframe). Yet, it suggests doing 10 problems daily for a month in order to complete 1000 problems. If your basic math is somewhat good, you will directly find the issue. ChatGPT has no idea of what it is saying or what you are asking for. It doesn’t have a good understanding of the world, but merely interprets the words and concepts of your questions and gives you back the most logical answers based on the different concepts in your question with no logic involved. Still, if you are just chatting with it and do not need applied and complicated advice, this chatbot is impressively good. It may be possible to fine-tune it to a sort of expert system manner with expert knowledge for specific applications to ensure the information it gives is correct, which wouldn’t be scalable to a general AI chatbot but might be interesting for a company-specific chatbot. This is a super exciting avenue with lots of potential!
- OpenAI released Text-Davinci-003
OpenAI says that it produces higher-quality writing, more complex instruction, and better long-form content. It is still trained on the same data as text-davinci-002 but tweaked differently.
- OpenAI released ChatGPT: a conversational-optimized version of the powerful GPT model
You must’ve seen the results online already. ChatGPT is kind of wild but obviously has some failure cases, as with all AIs. The dialogue format makes it possible for ChatGPT to answer follow-up questions, admit its mistakes, challenge incorrect premises, and reject inappropriate requests. You can try it here.
- Learn Prompting in partnership with Towards AI!
We are partnering with Learn Prompting in order to help build and spread how to do prompting and become better prompt engineers, which we believe will become more and more popular, and people will even be hired for this role in the near future. Check out our new discord channel for it and give them a follow on Twitter to see the new releases!
- AI Speeding Up Computer Graphics by more than 500%
NVIDIA’s Vice President of applied deep learning claimed “In certain GPU-heavy games, like the classic first-person platformer Portal, seven out of eight pixels on the screen are generated by a new machine-learning algorithm.” This is thanks to approaches like Instant NeRF released this year by NVIDIA.
Three 5-minute reads/videos to keep you learning
- How AI Understands Words
Large language models. You must’ve heard these words before. They represent a specific type of machine learning-based algorithms that understand and can generate language, a field often called natural language processing or NLP. You’ve certainly heard of the most known and powerful language model: GPT-3. GPT-3 understands language and generates language in return. But be careful here; it doesn’t really understand it. In fact, it’s far from understanding them. GPT-3 and other language-based models merely use what we call dictionaries of words to represent them as numbers, remember their positions in the sentence, and that’s it. Here we dive into those powerful machine learning models and try to understand what they see instead of words, called word (or text) embeddings.
- Can An AI Be Sentient?
Multiple perspectives on sentience and on the potential ethical implications of the rise of sentience including my contribution with the amazing Lauren Keegan on page 20 called “Going Beyond Sentience Towards Morally Responsible AI”.
- Diffusion models explained. How does OpenAI’s GLIDE work?
A great explained video by my friend Letitia Parcalabescu explaining what diffusion models are (the architecture behind the most recent text-to-image models like DALLE and Stable Diffusion). Letitia covers a lot of topics with lots of efforts to create very clear explanations and great animations in around 10–15 minutes. This is a great channel to keep learning efficiently that you should definitely follow!
Enjoy these papers and news summaries? Get a daily recap in your inbox!
The Learn AI Together Community section!
Meme of the week!
Featured community post from the Discord
“Right now it goes through some basics (Cot, 0 shot CoT, etc), some advanced applications (MRKL, ReAct), and a comparison of 15 different prompt engineering IDEs. I’m actively adding content on various topics atm.
I’m looking for criticism as well as suggested topics to cover, and maybe even some contributions 🙂”
Learn more about this project here and support our fantastic moderator!
AI poll of the week!
TAI Curated section
Towards AI Article of the week
With each release, OpenAI is reaching closer and closer to the rumored GPT-4 models. With every iteration, many lessons are learned, whether these are text, codex, InstructGPT, or ChatGPT models. Both performance and the safety of models are being improved. This article is about ChatGPT released by OpenAI. This model was trained to have interactions conversationally. According to the description on OpenAI, it is trained to follow instructions in a prompt and provide a detailed response.
Other must-read articles
If you are interested in publishing with Towards AI, check our guidelines and sign up. We will publish your work to our network if it meets our editorial policies and standards.