Yes i always advise to pick a domain, don't be a generic ML person. I've worked on Speech Synthesis for over a decade and everyone who ever reached out to me was because of this niche (at least it was a niche a decade ago, isn't anymore either). Nobody else ever cared about being able to stitch together some resnet in pytorch.

Similarly when we hired there were always hundreds of ... generic ML people but usually only a handful in speech and at best 1-2 TTS.

But then, nowadays people just tokenize everything and stuff it into transformers, domain knowledge used to be worth more ;)

Ah, I also started at 11ish and we got schools with technical focus here that you can go to at age 14. We had 7h programming hours per week and then tons of home exercises in C, Assembly, then C++... I am now over 40 and still love what I do. But the teens were the best time, everything's still so magical

Mich auch, feit Jahren

Hmm I also felt that some items in BG3 felt more... unique than in most other games. But I never enjoyed the whole looting and gear mechanisms in most games. I find it super annoying that in Diablos you spend so much time in your inventory and in most games I run with some crap all game long. I didn't finish a single set in Witcher 3 for example (luckily on death march it was still easy enough to not have to deal with it... or oils or elixirs or anything)

Ja, wir sind von Wien aufs Land und da sind die Interessen schon eher anders ;). Alle am Skifahren Berg kreun irgendwelche Dorffestln oder Kirchenzeug (ich bin immer noch schockiert wie stark der Einfluss der Kirche hier noch ist und der deppade Pfarrer immer überall mitzureden hat).

Aber zum Glück bin ich asozial genug dass es mir meistens reicht, virtuell mit Leuten zu verkehren. Naja und mit den Kindern genug ausgelastet.

Das wird viel zu viel schlecht geredet hier. In meinem Freundeskreis waren immer alle Partner auch dabei und sehr schnell integriert und fände es auch eher seltsam, wenn man mal was macht (was mit zunehmendem Alter sowieso immer weniger wird) und dann lässt man den Partner daheim hocken.

Der größte Freundeskreiskiller war allerdings als wir Kinder bekommen haben, und irgendwie sind wir jetzt über 40 und die anderen tun immer noch so als hätten sie eh noch ewig Zeit. Gut, sind auch hauptsächlich Männer aus meinem Kreis ;).

Insofern sind meine langjährigen Freunde jetzt mehr nur noch "Brieffreunde". Hin und wieder treffen wir uns mit anderen Pärchen mit Kindern aber brauchen sowieso schon alle monatelang Vorlaufzeit ;). Inklusive uns selbst.

Asking questions and then being a jerk about the answers also isn't exactly what brings you a good network.

It wasn't even mentioned it was the same job. For example I haven't even updated my CV in over a decade (tun fact it's been written using Netscape Composer originally) but have been in 3 regular jobs since then plus a dozen freelance projects plus a part-time teaching position. Never cold-applied, always been reached out to because of a mix of PhD Thesis, some paper I published, a GitHub project, a referral, acqui-hiring...

Yeah and then calling everyone else sour grapes lol

And I am pretty tired of people calling it a fad while having no idea what exists beside chatgpt. Ok we can discuss if you want to call "classic" (lol) deep learning AI or when people don't hype but just use it.

Some people already mentioned quite a few examples.... I've worked in ML for over a decade now (after another decade as regular dev) on medical computer vision, text to speech, speech recognition , video understanding and -search, predictive maintenance and other topics and the improvements we've seen have been massive. Multimodal foundation models are just getting started. With CLIP you can already search for "red shirt with a blue parrot on a palm tree" without the shop having to label every single attribute about their products. With newer video models a model can reason about sequences of actions in videos. The degree to which we already can index/embed, search, understand, transcribe, caption, translate etc. media is pretty crazy by itself.

Having started out in the medical field I am especially interested to see where AlphaFold and friends go. Friend of mine is working on smell (https://www.osmo.ai/) which is also pretty awesome.

I am in central Europe so things may be very different from wherever you are.

Well I checked various options at my universities but they were all rather unsatisfactory ;). Like 20h/week paid contracts limited to two years etc

I found a job ad for a position at a research center with expectation to do a PhD there and went for that. Had a few interviews, presented my master's thesis and was hired. Salary wasn't really worse than what I had before as it was a full, unlimited 40h contract and before I only worked for tiny companies. All in all it was pretty similar to a regular job except mostly having papers as goal instead of a product although usually 25% of the work time was supposed to be in projects for partners and clients. So while I worked on my PhD in Text To Speech synthesis I also hopped between other projects like in-car UIs, Citizen participation platforms, some weird large EU project (https://cordis.europa.eu/project/id/248567/results) and others. Later our group had enough funding on our own to have me full-time just on the topic.

Overall I liked it. I just realized I don't like writing papers and having that as a main goal so I soon decided to get out again after finishing. Directly after graduating I was snatched by a US startup in the field - it was still pretty small back then, a handful of research groups worldwide, knowing each other.

After a total of over a decade in that field I had the chance to at last get to work on something else and that's what I am doing now (multimodal model stuff mostly... Video understanding, vid o search, things like that, of course also have to do LLM and RAG stuff because everyone does atm). Once you're seen as an expert in a specific topic, changing to something else becomes pretty tricky so a company-internal move was a pretty good option.

I see what you mean but I for example am very product-driven. And the product of LC is artificial and sandboxed at best.

But I had the same so after a few years as dev I did a PhD and since then have been working in applied Scientist positions and can't complain anymore

Ugh that's of course much worse, we only had 3 "talk to each other and present your work" style interviews so it wasn't too much time wasted for the candidate.

But I as hiring manager in total definitely spent at least 40 hours till now, the recruiter talked to some 40 people, my team talked to 4-5 people...

Yeah.. I've interviewed tons of candidates, we found one we really liked, everyone agreed.

Then my manager interviewed her, said he wasn't impressed and he'll talk to her again at another point in time.

2 weeks later they said we're basically pausing hiring till next year, at least "re-evaluating all open positions".

I feel like an idiot going out to tell her that... well I guess it should be the job of the recruiter but I think nobody will do anything if I don't do it. And that's not fair after we all said how happy we were with her performance etc.

It just sucks.

Well but at least we're not a Google making 20B net income a quarter while still doing this, we really fight the tough economy. Still sucks

Master's is interesting if you want in a specific field where you'd otherwise have to spend a lot of time skilling up yourself anyways. For example computer graphics at TU Vienna is quite renowned and gives you access to a lot of industry contacts if you're doing well.

I did my master's at a medical university for medical informatics and got direct insight into all kinds of cool equipment like high power MRIs and for my thesis worked on image registration for laser treatment of wet macular degeneration. The people I now know working there were all directly recruited from the students.

Also the CVs that appeal to recruiters the most are likely not the best developers but those who have to spend a lot of time polishing it. Many really good people I know haven't updated their CVs in a decade because they've been informally snatched all the time.

Of course there will always be someone great in there as well just deciding to shop around, being affected by a large spontaneous layoff or never were able to make their contributions visible enough.

But generally the better people are at interviewing probably the more they had to train for it

Ironically the most polished CVs are also likely from people who have to work hard on their CV. Most really good people I know haven't updated their CV in a decade because they were just grabbed by... word of mouth

Frag ich mich auch oft. Mit dem Alter wird's definitiv blöder. Beim Zivi waren wir gezwungen, den ganzen Sommer lange Hemden, lange Hosen und die dicken Arbeitsschuhe anzuhaben und die Klima in Auto ging auch oft nicht oder es gab keine. Da dann teilweise auch Mal schnell eine Stunde irgendwo auf einer Straße gehockt oder Leute irgendwelche Böschungen hochgetragen.

Danach manchmal die Handschuhe umgedreht und ein ganzer Schwall Schweiß rausgekommen ;).

Aber da war das mehr lästig weil ständig das ganze Gewand komplett waschlnass vor Schweiß. Fertiggemacht hat es mich nicht.

Heutzutage setze ich 20 Minuten nichts auf in der Sonne und hab am nächsten Tag sicher Fieber.

Sind in eine kühlere Region gezogen aber denke noch oft daran, wie es wohl in unserer ehemaligen Dachgeschosswohnung in Wien läuft.

Which makes sense for the FPÖ. Their targeted demographics are not business owners like the ÒVP or Neos. They only advertise for less governmental influence when it pleases their crowd, and that's more about ... COVID, speed limits, "WHO Wahnsinn".

I mean obviously there are tons of alternatives like "Löffel abgeben" or "abkratzen", "verrecken", "draufgehen", "abmurksen", "kaltmachen".

Having lived long around the region of Vienna with its dark, colorful language I could probably come up with many others... Patschn strecken (botschn streckn), einen Schuh machen (an schuach mochn), abbankeln (obangln), jemanden heimdrehen (an hamdrahn), jemanden abmaxln (omaxln), abfeidln...

But I would not know of any censorship.

Wild, we got 9M (Austria) and over 500k people are in the Austria subreddit alone ;)

You are right, finishing with enthusiasm can be super important. We rejected a Harvard Physics PhD because it was so obvious he didn't care about the topics of the job at all. He just answered yes/no on everything. That also explains to me why some interviewers always ask "tell me about a time where you worked with X" instead of "Do you have experience with X?" and assuming they would elaborate instead of just saying yes or no ;). It was almost impossible to get any discussion started except when we asked about his hobby passion, then he didn't stop talking (but that wasn't related to CS at all).

At the same time we all agreed upon a person who had a different background but was obviously so enthusiastic about the topic she'd be working on, gave a super polished presentation about a hobby project she did and ended with how interested she'd be in this job. And it really seemed sincere.

My actual favorite was a guy from ByteDance who I had really deep conversations about our topic with and it also goes the other way round - he wanted to continue the process even though it became clear that he probably earns more than 2x or 3x (seems the recruiter wasn't clear about base salary vs total comp...) because he enjoyed our discussion so much lol.

Still feeling bad about for the people though because after we put so much effort into interviews "all open positions were re-evaluated"

Ironically it's what I dislike about security - not improving the world but just dealing with idiots ruining it. I'd rather build stuff.

So during my civil service that I do instead of our compulsory military service I worked a year as a medic for the red cross and enjoyed it that I afterwards studied medical informatics. Strictly speaking even before I did that it didn't feel useless for people - I've worked on embedded systems that were monitoring energy consumption and also aspects like temperature, humidity, door contacts etc. So you could alarm when someone entered a room and didn't leave anymore, monitored machines so they don't burn down and cause havoc ;). Anyways, I then studied and did my master's thesis at the Opthalmology in a big European hospital, topic was computer vision for laser treatment of wet macular degeneration. I then did a PhD in speech technology, we.worked with blind school children, tried to improve their technology. I've been at their school and a few days at a summer camp. It was a cool topic. We also initiated a project about sign language but didn't play out with the funding.

After that I was hired by a startup on voice banking, providing TTS voices for people losing their voice (head-neck cancer or ALS for example, also many kids).

So it was quite a journey but honestly meanwhile I often think why I am doing this. If I look at all my friends in healthcare how much hate they get... Especially since COVID, they've been physically attacked and the hospital walls have been sprayed with hate messages. People rather believe some idiot on YouTube or TikTok because "they all get brainwashed from pharma at university". Even my wife how studied veterinary medicine got random people coming up and starting discussion about how she's paid by pharma to be a proponent for vaccinating their dogs. They use silver or lemon juice or whatever instead.

And the reality is, you typically earn much less in almost any job that's useful for society. Be a parasite HFT and get all the money instead. I am now in a job that pays 4x and I often deal with... influencer crap and... all kinds of other crap but honestly I really feel at this point I have to do the best for me and my family and not all those idiots out there who don't deserve any better.

Luckily I finished my PhD before things became so crazy.

But honestly now I am in industry and have exactly the same looking at LinkedIn or any news feed.

3 new products/startups competing with you, 7 new models competing with your offering. We have a very good customer retention but it's such a rat race anyway. "Oh but your competitor already got a cool copilot built in, oh there's a better multimodal search offered by X, oh your cool analytics model grabbing from Salesforce is now a Salesforce feature, oh Azure offers this a lot cheaper now"

Sometimes I wish I could just happily work on a product that's just ... stable and you know, you just work on it without everything always being deprecated in 4 months again because either the state of the art changed or it gets stomped because everyone's flocking to the new.. ElevenLabs, TwelveLabs, whatever lol.

But I guess in this field it's almost impossible if you're not in a super specific niche. Idk, people at Mistral also pissed at every Antrophic release? ;)

I've seen worse but around this is where I dig around regularly and it's... Challenging ;) https://github.com/espeak-ng/espeak-ng/blob/063e3962dfc108db4c8b6fd05f4a9b531d74dd2f/src/libespeak-ng/dictionary.c#L1450

(To add: this codebase is decades old and grew.and grew so no blame from me)

Wow, had to re-read my own answer ;).

No, I just wanted to say that I don't see a shortage in backend developers at my company and no trouble hiring either. And they somehow see us 4-5 ML people as rare. While in reality we get so, so many really good applicants for any ML position we offer.

This does make me a bit nervous honestly, but well... ;)

(Regarding security I saw this posted here recently https://cyberisfull.com/ )