Animesh Sinha is a research engineer at Meta AI. He is a technologist with expertise in computer vision and multimodal understanding (vision & language. Currently he is focusing on Generative AI.
To stay in the loop, we have put together the viewpoint of Animesh that you could follow to know the progress brought by Artificial Intelligence in the tech industry.
Let us quickly get to our expert's point of view.
Ans : I am very optimistic about generative models and scaling up models in general. Generative models have become pretty good at generating data (text and images) that they could easily be used to solve the data sparsity problems, which can result in much better models across the board. We all know that most AI systems get better with more data, so these generative models can be used to generate more data in domains where we lack real world samples.
For example, we have seen that models like ChatGPT or Stable Diffusion produce biased results, most of those issues come from a biased and unbalanced training dataset. These problems can be mitigated to a large extent if we have a system to balance it. In terms of data, another example where these generative models could be highly useful is autonomous vehicles, where they can be used to generate samples or environments which occur very rarely in real life, thus creating richer and more diverse datasets that can prevent a lot of accidents or errors from happening in production.
Scaling is another promising research direction that a lot of labs are looking into at the moment. We have seen new behaviors emerge from models when they are scaled to billions of data samples or are scaled to billions and trillions of parameters. I think the models will keep on getting bigger and better and will open new experiences that weren’t possible before.
Overall, I think the current AI systems can have a great impact in sectors like health care through early diagnosis of diseases, personal finance, medicine through drug discovery and transportation through self-driving vehicles.
Ans : There are multiple areas where one can leverage AI in 2023. Some of the ones top of my mind are:
Data : Data will always be a very crucial piece to improving AI over time. AI could be leveraged to better curate existing data and create new data so that we make data more easily accessible to people.
Customer Support : The large language models have started working well enough that one could repurpose them to improve one of the biggest problems in customer support.
Marketing : AI can be leveraged to help SMBs with marketing campaigns (a good example is the Cadbury ad in partnership with Rephrase AI) and deliver a personalized experience to the customers.
In general, we will see many new use-cases of “foundational models” where they will be repurposed to solve smaller and specific tasks.
The last event I attended was a small meetup of founders in San Francisco to know how they are using AI and what new business use-cases they are solving. Some of the key takeaways were:
It’s still hard to deploy big models in production, so people are working hard on model optimization and improving model serving pipelines
It’s still hard to deploy big models in production, so people are working hard on model optimization and improving model serving pipelines
People are thinking about very unique product experiences that could only be possible due to the current large models. It’s a very exciting time in AI because the field is moving so fast and new things are becoming possible every few months.
People expect AI to solve everything out of the box, which is not true at the moment. AI is still in a nascent stage and requires a lot of failsafe in place when launching in a real life product
Ans : The USA is one of the leaders in pushing the state of AI forward. There are many companies working at the cutting edge of AI and are already testing it in real products. There is a lot of potential for AI in sectors like climate, medicine and transportation, and the USA can take a lead in showing how to use it in a constructive way and set the right example.
Ans : AI has already had a major impact on supporting remote work in the last couple years. From applications like virtual background, improving video quality, taking meeting notes automatically, AI is everywhere. We are not that far away when people won’t need a physical laptop and keyboard, and will still be able to work seamlessly.
On a similar note, Microsoft’s partnership with OpenAI is also going to make work much easier with the help of an AI Assistant in all of the MS Office offerings.
A lot of the companies are pivoting to become AI first companies by integrating AI in all their products. We are also seeing a huge rise in the number of AI startups in recent years, so AI is having a profound effect on new and old companies and this trend is expected to continue in the years to come.
Ans : I don’t think there is any other way to be successful than putting in a lot of hard work, failing and learning from them and not giving up. I try to constantly learn from what worked and what didn’t, and try not to make the same mistakes again. I also totally believe in what Jack Ma said once - “Today is difficult, tomorrow is much more difficult, but the day after tomorrow is beautiful”.
Senior Consultant
Leave your details so we can contact you back.
What to expect from AI in 2023
AI is emerging and evolving faster than ever. AI is entering in a promising era, beginning in 2023, thanks to vast improvements in processing power and computational efficiency.",
Animesh Sinha
