FlyingSaucer

Maybe its only my gig and my area of the world, but going through the rounds with companies can be quite a grueling process and many recruiters can't explain the exact technical details or give any concrete salary information.

It's hard to just do it to test the waters when you have a technical project and multiple interviewing rounds (which also include live coding, which for me is very stressful).

I'm by no means an expert in the field but I do find it exceptionally interesting so i try and keep tabs on some of the research done by people who originated from the same group as Kenneth Stanley.

Seems like many from this group now pursue open-endedness in AI and view evolution as a way towards this goal (or lack thereof).

A very interesting evolution (ha!) of these ideas was presented in POET[0] towards evolution of agents in evolving environments.

There is also an interesting paper about accelerating neural architecture search when generating fake training data in generative teacher networks[1].

Lastly, a paper that i find very very interesting but might not be as relevant but still is 'First return, then explore'[2]

[0] : https://eng.uber.com/poet-open-ended-deep-learning/

[1] : http://proceedings.mlr.press/v119/such20a.html

[2] : https://arxiv.org/pdf/2004.12919.pdf

I completely agree. I used to live in Liege and would travel to NL, Brussels and Hasselt regularly. The juggle between different public transport subscription was difficult on top of the constant outages.

The extensive bushfires and droughts are definitely big factors. I worked briefly for a forestry in rural Victoria, where we would encounter koalas somewhat frequently where we would plant trees, These are lots that would be planted completely and years later cut and burned.

Wonder what would be also the effects of this on them and wildlife in general.When I have seen Koalas in the wild they seemed very apathetic- saw one stay at the very top of a blue-gum (which famously can shed huge pieces without much warning) even during a storm.

I have used the MLP classifier[1] before. It's very simple to use (like most of sklearn's models). Worked well for standard and reasonably small classification model, but lacks some features for it to be a flexible way of using NNs:

- No saving checkpoints (can be crucial for large models who need alot of compute and time)

- No way to assign different activation functions to different layers

- No complex nodes like LSTM, GRU - No way to implement complex architectures like transformers, encoders etc

I also do not know if its even possible to use CUDA or any GPU with it.

[1] : https://scikit-learn.org/stable/modules/generated/sklearn.ne...

from the article: 'Warning sign (of a critical moment in AGI development) will be, when systems become capable of self-improvement.'

Is the meaning of self improvement here means that the model will actively optimize itself towards improving on its mistakes outside of training? Because under my understanding for this to happen we would need the model to be in a different form than current ML.

While many units generally consist of smaller teams, the teams can often be changed and switched around as people get specialized or re-assigned at different points, this generally doesn't happen for the very elite units.

Small teams are kept together at all times since very early in the training. Doing the same exercises many many times with the same exact people leads to mastery that makes them just seem faster and more fluid.

The lifestyle point is also true, its easier to keep your edge when there aren't constant mindless tasks to be done (gate duty) for many hours a day.

- I'm saying this from experience with armed forces, but can't claim its true for all elite units everywhere.

This is from my own experience working at a TP call center, for a major tech client and not in Albania or Colombia.

External and internal audits are known for weeks in advance such that they could put a facade on in time. During audits you suddenly don't have a limit on bathroom break time (generally an alert would come at 15mins per day) and there would be pizza at the office every other day. Only 'loyal' employees would be chosen to any interviews/meetings about culture etc.

Yes, its hard to tell the exact algorithmic underpinnings of production models that Google uses but you have to assume that although they have some done some impressive strides in fields that isn't immediately profitable (AlphaGo, AlphaFold...) they also continuously push new research in things that are obviously of interest for Google and Alphabet- especially in text-to-speech, speech-to-text, information-retrieval etc.

For reference : https://deepmind.com/research

Yes! The subtitles can be completely different than the dubbing in some cases which makes it difficult for me to watch with both.

Also, they sometime use a bit odd translations. I saw a Belgian show and they translated smoking a cigarette into smoking a fag. Which I guess is technically correct (based on the cambridge dictionary), just an odd choice for general EU viewership

English isn't my native tongue so perhaps i'm missing some subtlety, i thought ethnic groups are people who share a common descent?

Honestly asking, not trying to nitpick.

Huh, so its not only me. My grandparents came from Libya in the late 1950's, but for some reason the very north of Africa doesn't quite qualify for most people as African "enough" for me to be of African descent. Feels like for many people the benchmark is either skin color or not coming from a traditionally Arab country. A similar thing happens to acquaintances of mine of Tunisian or Algerian descent.

I guess the geographical line is somewhere around Mauritania

Depop isn't alone in this, Vinted[1] has a current valuation of $4.5B and is hugely popular in Belgium, Netherlands and France (maybe more places, but those i can definitely say).

In both platforms there is influencer-heavy marketing. Many fashion influencers(i really dislike this term, but it is the nomenclature) sell clothes that they wore very little in Instagram advertisements. This is a great business model for them, people follow them more tightly because they are actually able to buy for (mostly) affordable prices some of the clothes they see in pictures and they also get some additional under-the-radar cash.

[1] : https://www.vinted.com/

That's the biggest annoyance in my eyes. As a motorcyclist, i think its ok for the engine to make some noise (not counting the obnoxious cruisers or bike with some aftermarket exhausts), but there is truly no reason to rev up hard while sitting in traffic or in a slow central streets.

The name of the actual study is 'There Is No Evidence That Associations Between Adolescents’ Digital Technology Engagement and Mental Health Problems Have Increased'[1].

I also personally feel like its incorrect based on my own experience, but the OII for now says that there is 'little evidence', although 'drawing firm conclusions about changes in their associations with mental health may be premature'

[1]-https://journals.sagepub.com/doi/10.1177/2167702621994549

I'm a DS by title. I don't do any dashboards. Some people on the team collaborate on them occasionally but its generally deciding which metrics are tracked along with creating/maintaining them while the actual dashboard design and creation is done by the frontend team.

The thing is that Data department have essentially swallowed the former analytics departments, and many people who have done business intelligence/business analytics now seem to fall into the data science umbrella.

This is part of the reason the term now refers to different things depending on who you talk to or which team they are a part of. Look at job postings, its extremely difficult to understand what the actual job entails these days.

I don't see it in this way. Pierre (which is the owner of the local general store) is by no means a pure character- he will take credit for produce that you sell him and will generally sell with a very high margin compared to how much he paid you.

Now... are you the villain if the junimos are the ones doing all the work and you just reap the benefits?

I think i agree with your general sentiment, but specifically LSTMs were first proposed in a 1997 paper[1], thus my guess is that they didn't have vast resources compared to today. I mean, i trained my model on a GPU for a few days, which im guessing is actually more resources than they had.

>Few architectures are designed or evaluated based on smaller (fixed) corpora sizes and smaller (fixed) training budgets. Even few-shot learning tasks typically still require a huge amount of pre-training on large datasets. So researchers and practitioners constrained by fewer resources and smaller datasets (which may not apply to you specifically) trying to adapt popular architectures to their needs are disadvantaged. Compare the attention being given to energy budgets and similar constraints for inference as opposed to training and the disparity becomes fairly obvious.

that is an interesting point, and i feel like it generalizes to the fact that using more efficient architecture that was perhaps was designed by someone with a lesser training budget. Although I must say that from my limited DL paper reading, efficient small-scale novel architecture doesn't necessarily comes from cash-strapped researchers, as a more efficient(energy and time) would be of huge economic value also to companies like OpenAI, who have spent huge amounts on training GPTs.

[1] - https://www.bioinf.jku.at/publications/older/2604.pdf

> Fixing racial bias in AI is not just a matter of infusing the training data with more melanin (for example), the AI Ethics crowd argues — the actual models are being developed by white guys, and their insular, white-guy priorities somehow surface as bias in the algorithms that go to work on the training data.

I recently created a natural language generation model(built with LSTM layers mostly) that was trained on east-asian zen books. Do you think that my result could have been better if I would've used an architecture not designed by white germans?

This idea seems is to me like anthropomorphizing model architecture for no real reason.

I do feel like there might be issues of ingrained bias in the model itself when using trained NLP embeddings or even some facial feature recognition algorithms that were tested on racially homogeneous groups.

I'm not OP, but as someone who traveled for a few years without the goal of 'finding myself' i actually found the experience pretty transformative to my character. Yes, Some of the more 'backpacky' memories do fade(might also have to do with the excessive drinking associated with it).

I still found that by immersing yourself in other cultures and ways of living you can gain more perspective about your own goals in life and your personal choices.

Interesting to hear, in Belgium for example the homeownership rate seems high(speaking from living here for a while). The culture is very much geared towards buying a house, alone or with a spouse. It's also a small country where people don't move alot, which might mean that they don't value flexibility as much as larger countries where moving is generally over larger distances.

Also, the banks offer very good rates for these types of loans if you have a stable job and prospect.

Renting for very long periods seem like a waste for many people here.

This topic is extremely interesting, and good to see that experiments support it.

In AI lens:

In a way, you can compare this to novelty seeking and intelligent exploration which is quite an active field in Artificial Life and game AI[1]. If you find this interesting: Jeff Clune, Kenneth Stanley and Joel Lehman conducted interesting related research.

Also, isn't this somehow related to the Free Energy principle by Karl Friston? If you look at entropy maximization as a way to minimize surprises.

[1] : https://arxiv.org/abs/1901.10995