nikvaes

  Location: Amsterdam
  Remote: open to remote (or hybrid in NL)
  Willing to relocate: within NL if needed
  Technologies: java, typescript, python, pytorch, cuda, SLURM, terraform
  Résumé/CV: available on request
  Email: username at google's email service

I have a PhD in speech recognition and a few years of industry experience. I'm looking for any roles touching AI systems (research/engineering). I'm happy to work on features end-to-end, my main motivator is building software people love to use.

Some previous work:

- trained speech foundation models

- worked on a spoken dialogue system for drive-thru's

- worked on a mobile fintech application

- worked on Jitsi Meet

- written a popular python package (https://github.com/jitsi/jiwer)

It feels like this is an advertisement or data collection effort for your platform, as after the application, you get redirected to your product, which never even sends an email?

Bus lanes make a bus a very good alternative to driving, and they are much more efficient space-wise compared to cars (and god forbid, pick-up trucks), which means you can get more throughput...

they already do that, it's called SURF, and they host a nextcloud instance called surfdrive.

For people who are interested in the question whether self-driving cars are the solution to commuting: an video essay of why self-driving cars are worse for a city than public transport: https://youtu.be/040ejWnFkj0?t=1684

As I had the same feeling as you, I subscribed to a quarterly news magazine called delayed gratification. I feel it's a good balance between keeping up-to-date, while not letting the news interfere with my daily life and emotions.

https://en.wikipedia.org/wiki/Flevoland used to be (part of) a sea.

It seems this is referring to https://beviai.com/

I think it is good that tax money is spend on having a well-educated population, this is something a democracy thrives on. It is a shame it is spend on loan forgiveness instead of making education cheaper so loans are not needed. Loan forgiveness should be paired with reform so future generations do not have this problem.

Isn't consent by default illegal under the GDPR?

I just want to point out that I think ChatGPT has seen more data than a single human being. With 45 TB of text data [1], it would cost a human around 70k years of continuous reading (assuming 1 GB is 200M words, and humans read 225 words per minute) to see all the training data.

[1] https://www.businessinsider.com/google-researchers-openai-ch...

After trying to understand and implement some algorithms in RASP [1, 2], my take-way was that certain functions need a certain amount of transformer layers to operate. Following this logic, it should become apparent that the functions learned by transformers can be spread over multiple heads. Repeating these functions might be very valuable for understanding and solving a problem, but current inference does not allow (a set of subsequent) heads to be repeated. This paper indeed seems a promising direction.

[1] https://arxiv.org/pdf/2106.06981.pdf

[2] https://www.youtube.com/watch?v=t5LjgczaS80

Did you also look at the false positives, e.g., how many non-spam content was filtered by Akismet?

In my opinion, a first step is to equalize maternity and paternity leave. It should be equally disrupting for men to have children as for women. I like the scandivian implementation of this model, where partners get 2 years of leave per child which they are free to divide among themselves, with a minimum of a few (?) months for either.

Yes, they have registered a company in The Netherlands:

https://support.discord.com/hc/en-us/articles/4410339409047-...

https://www.kvk.nl/orderstraat/product-kiezen/?kvknummer=822...

The problem for Jigasi's speech-to-text feature with Whisper - or any recent SOTA speech-to-text neural networks, is that they are transformer-based. One of the key features of transformers is that they are very good at processing a sequence with the attention mechanism. But attention inherently needs to see the whole input sequence. So it's difficult to adapt these architectures to perform well in real-time scenarios like captioning meetings.