We have entered the era of generative artificial intelligence, which has manifested itself in various applications, ranging from the composition of essays and creation of comics to the editing of films. The adoption rate of generative AI has surpassed that of any other consumer technology trend observed in the past ten years. The text generator ChatGPT, for instance, accumulated over one million users within a mere five-day period, while tens of millions of consumers have generated AI avatars.
The majority of venture capitalists regard generative AI as the forthcoming primary platform upon which entrepreneurs will establish groundbreaking products. At present, large language models (LLMs) demonstrate proficiency in addressing challenges pertaining to problem-solving. Consequently, even a simplistic game like Flappy Bird can be developed within a few hours. The capabilities of amateur programmers are significantly amplified and augmented tenfold through the utilization of generative AI.
In the foreseeable future, it is plausible that average users will experience substantial empowerment in the realm of software development. Extrapolating this trend, we may anticipate an era in approximately a decade where average users possess the ability to construct intricate software applications.
However, for the creation of serious applications, it is imperative that these models exhibit reliable behavior. Moreover, a mechanism must be established to facilitate the provision of concrete requirements as input. Finally, it is essential to eliminate any surprises or unforeseen behavior in the output generated by these models, thereby ensuring the dependability and usability of the software developed using generative AI.
- 100% reliability and edge cases. e.g Tesla example (need to think of alternative approach)
- Can’t solve reversal problems in healthcare - requires argumentation
- Can’t edit model
- Alignment (function. And RL) problems
- How do we query the LLMS in a privacy preserving way ?
- how do we do in a compute efficient way, how do we run it at the edge. how can we make your fridge a smart ridge, who knows about your nutrition intake and so on, without actually being connected to the internet. ?
- Zeno: An Interactive Framework for Behavioral Evaluation of Machine Learning
- Emergent Abilities of Large Language Models
- [GPT-3: Its Nature, Scope, Limits, and Consequences]
- [Sparks of Early AGI]
- Emergent autonomous scientific research capabilities of large language models
- GPT-4 Has the Memory of a Goldfish
- ChemCrow: Augmenting large-language models with chemistry tools
- OpenAGI: When LLM Meets Domain Experts
- The Logical Structure of the World (1928) - Carnap’s book defined an explicit computational procedure for extracting knowledge from elementary experiences
- Kangas 2.0: Exploratory Data Analysis for Computer Vision
- A Statistical Data Testing Toolkit
- Converting Tabular Dataset(CSV file ) to Graph Dataset with Pytorch Geometric
- Zeno: An Interactive Framework for Behavioral Evaluation of Machine Learning
https://github.com/dair-ai/ML-Papers-of-the-Week#top-ml-papers-of-the-week-april-3---april-9 https://www.promptingguide.ai/papers
- Knowledge retrieval for internal documents - openPilot
- https://www.aitoolsclub.com/