Using AI to solve complex problems | Gemini

Watch as AI solves complex problems with Google DeepMind Research Engineer Gabriela Surita – using Gemini’s advanced coding abilities, including quickly prototyping a web app to explore London’s train stations. Gemini is Google’s newest and most capable AI model. Research Scientist Rémi Leblond also introduces AlphaCode 2, an advanced code generation system that excels at solving competitive programming problems involving complex math and theoretical computer science.

[kinlearn]

Waiting a crisp "Code Report" from Fireship 🔥

[caelen_c]

Can't wait to use this in my projects

[QueenLover-ji]

- Understand Gemini's capabilities ([0:09])

- Recognize improvement over PaLM 2 models ([0:21])

- Consider Gemini's self-checking feature ([0:36])

- Explore AlphaCode2's performance in competitive programming ([1:55])

- Evaluate the use of dynamic programming by AlphaCode2 ([2:57])

- Acknowledge the collaboration between AI and human coders ([4:06])

[craftmechanics]

The problem showed in the video is from CodeTON Round 4 which was held 8 months ago. There are video editorials on YouTube explaining it which are most definitely in its training data.

They need to test these models with original problems which don't already have solutions in the training data...

[saiphaneeshk.h.]

This is what I wanted, let's hope they have gotten better in helping to code more efficiently and accurately

[franciscogeiman]

It would be nice to have a codeforces rating for the bot, maybe making it an official participant in contents from now on?

[sauravroy]

Well, Its official, The ride is over boys. It was fun While it Lasted. I'm gonna move to the woods in a log cabin with four hens and a Cow.

~ Indian coder

[haroldpierre]

Although I am not a programmer, I have utilized GPT-4 for my projects in Google Apps Script. GPT-4 is not without its flaws and I will get stuck from time to time. I am hopeful that Gemini will address and rectify the issues I've encountered with GPT-4.

[Shbn]

Imagine how beneficial would be a more generalized version of Gemini to individuals with vision impairment. This may open the door and allow a new generation of biomedical devices that enable them to experience a sense of spatial awareness by navigating their surroundings with unprecedented autonomy (guided by a continuous stream of auditory information about obstacles, potential hazards, and so on).

[sanjayanps]

Future of programming for about 3-4 months - yes.

[RAMKUMARRAMESHR]

Thank you, Google, for enabling this achievement. It brings me great joy to share this success with everyone involved.

[MEGIDIOT]

These two engineers are working towards making themselves unemployed

[wonerkang]

One of the problems with using GPT-4 is that the methods are limited. If you ask it to search for something that exists in the real world and create an image of it, it will always stay in the mold you created before and repeat your mistakes. Make a small improvement and GPT-4 moves on to another chapter, or make a small improvement and suddenly the paradigm shifts. Gemini hopes to avoid repeating these GPT-4 mistakes. I think Google has an advantage because they have years of experience data from tons of people, but as they say, with great expectations come great disappointments, so I will always look at GEMINI with a filter of caution.

[priyankkumar]

What will be the accountability and ownership for the code generated by Gemini ? Who would be the actual owner ? The person writing the prompt or Google ?
Also, Is it plagiarism-proof ?

[torarinvik]

I am really excited about this. Hope it really does outperform GPT-4 turbo in coding. GPT-4 turbo does not always do a good job, and has a tendency to make small mistakes, just like us humans.

[Billythek]

When I want to go back again and then the algo just gave me this. Perfect!

[yaragonak]

Gabriela Surita, I admire her so much!

[Kaishk]

too bad that competitive programming is the wrong thing to optimize a general programming model for - 99.9% of real-world problems aren't close to what you see in competitive programming.

[Yulenka-]

I wonder how good is this model at understanding various APIs the programmers code against, not just pure algorithms in a vacuum. This is a crucial factor for its application in practice to solving practical problems 🤔

[xtam]

Jst asked it to tell me how to update a jsonb field in psql, jst gave it database structure wid one data sample nd it said it doesnt have enough info, threw same prompt in gpt 3.5, nd it gave perfect response in one try

Man u guys really need to improve urself