Minutes of the SIPB Meeting of 2023-01-09


The meeting was called to order at 19:30 by jmvidal.

In attendance were

        Student keyholders: cjq, huydai, jmvidal, toomas, xy

        Associate keyholders: agrebe, jeffery, kenta, markchil, medesaro

        Members: almonds, alwinfy, colclark, fabianv, guoj1112, rgabriel, roseh, shreyach

        Guests: '()

Administrivia:

        '()

Project Reports:

        cjq: iap is happening! javsolis isn't here but not it's happening, check sipb.mit.edu/iap. Speaking of the Cluedumps, it is traditional for the Cluedumps to have snacks of sorts. Does anyone want to volunteer getting snacks and bringing them to the Cluedumps?
        
        cjq: Great, Clark says yes. I want to propose an allocation of $50 for snacks for this Cluedumps.
        
        huydai: Seconded
        
        huydai: If you don't want to go too far, La Verde's is a great place.
        
        jmvidal: Motion to white ballot.
        
        (passes)
        
        cjq: Anthony and Javier is repeating their well-received Command Line Fundamentals next Wednesday at 4-231.
        
        Anthony: This is teaching to learn how to use basic command line utilities, connecting to remote servers, installing packages etc.
        
        cjq: I will co-teaching on designing good programs with types. Types, hooray!
        
        toomas: What is a type?
        
        cjq: You can be type, that's a type.
        
        jmvidal: That sounds like my type of course.
        
        toomas: What, Air Force?
        
        cjq: DCAI is also happening. 
        
        guoj1123: What is DCAI?
        
        cjq: It's Data-Centered AI, starting next week. If you're interested in Machine Learning, there is a class next Tuesday!
        
        jmvidal: also worth mentioning. if there's anyone interested in the supercloud, there's an email that huy sent a few minutes ago before the meeting. if you're interested in that, join the orcd channel in the mattermost.
        
        huydai: to summarize, they're creating a research computing virtual desktop, some platform where there's easy access to a lot of local and remote cloud resources, putting them together and making it easy to use things across multiple computing platforms. if anyone's interested respond to the email and join the Mattermost!.
        
        kenta: I would like to provide two cent plug for that. I was around for the last centralized computing iteration of this project called "longjobs". i was a grad student mentoring an meng student, she also did a bunch of compute on longjobs and completed an meng through it. having this kind of resource is useful. after longjobs shut down i personally ended up using sipb office heads for personal compute needs, but having something generally available to mit seems very useful.
        
        rgabriel: uplink continues. i'm working on moira list integration and debating whether to use python or javascript.
        
        cjq: just use python.
        
        toomas: As someone who has done some Javascript, I'm concerned
        
        rgrabriel: Actually Typescript
        
        toomas: Python is better (regardless)
        
        cjq: there will be more people who know python in the future.
        
        colclark: have you noticed that hydrant and courseroad are kinda dependent on each other? i'd like to organize weekly meetings for development because they need to be more connected! if anyone wants to rollup i'll make a mattermost channel. i'll put a when2meet in the hydrant and courseroad channels.

Other:

        cjq: The Stud is opened again. I've been hanging around in the SIPB Office the last few days, and will be here the next few. You can always check doorpi.mit.edu to see if the office is opened and come by, and eat the snacks, or something.
        
        toomas: how long has doorpi been operational?
        
        cjq: several weeks now.
        
        alwinfy: it might be down again?
        
        cjq: there's no certs, you have to do it on http, not https.
        
        jmvidal: It's up for me
        
        *several other people*: It don't work??
        
        cjq: Point is, come hang out in the office. Eat some food! There's also the computer.
        
        toomas: what games are on it?
        
        cjq: i dunno, beat saber at least because there's a whole vr setup.
        
        agrebe: Is the SIPB meeting next week in-person or remote?
        
        cjq: I will try to be here.
        
        agrebe: Is Stud opened on MLK day?
        
        cjq: Yes.
        
        cjq: On the topic of Wifi and computer networks, IS&T is deploying an update to MIT and MIT GUEST on Friday morning. I think there was a whole announcement on their website on what is happening?
        
        agrebe: my understanding is that you have to get some password ahead of time for MIT, and some portal for MIT GUEST, you can no longer autoconnect devices: <https://ist.mit.edu/news/open-wifi-changes>
        
        guoj1112: (reading) MIT is moving network to one that requires Preshared Key (PSK). User will need to visit the IS&T new network portal to get their key at <https://wifi.mit.edu/>. You can fetch your key in advance. The MIT and MIT Guest network will provide a captive portal. And MIT Secure will stay the same.
        
        toomas: because it's already secure.
        
        cjq: So true bestie!
        
        huydai: as a security person i thought, this is a really good change. i can already imagine that wifi-connected lightbulbs, whatever won't work.
        
        cjq: If you go to wifi.mit.edu, you can request a Wifi password that lasts for 5 years. I *think* it is one time password (I hope).
        
        kenta: If I understand correctly about the Wifi changes, it means that MIT will have an identity associated with each packet that is sent over. So there is less anonymity with the wifi traffic
        
        colclark: i can request to reset my password and it resends it in plaintext, so they store it in plaintext.
        
        cjq: Yikes!
        
        guoj1112: The LastPass hack continues to be a problem. i don't use it but idk if anyone else does.
        
        cjq: it is IS&T's recommended password manager.
        
        huydai: i use keepassxc.
        
        almonds: same.

Other Other:

        huydai: apparently microsoft is incorporting chatgpt into bing? the demos look cool.
        
        alwinfy: that's gonna go well....
        
        roseh: they’re also planning to integrate it with office, generating filler text and stuff
        
        toomas: reminds me of that microsoft ai that learned to be racist.
        
        guoj1112: But that was 7 years ago.
        
        almonds: there's been a cambrian explosion of web extensions that hookup to chatgpt. every day you see they're scaling up. it's ridiculous how many people are using it.
        
        guoj1112: Brazil had a bad day yesterday.
        
        cjq: isn't your whole cluedump on wednesday about LLMs and their implications
        
        almonds: yes that is the title. i guess it doesn't say chatgpt in the title or the description
        
        medesaro: I am working on a project to decode old analog video off of video tapes. This involves processing ~200GB raw data files. With my current desktop PC, I am taking 20 hours processor time for every hour of video. I am debating buying a new computer or buying Amazon time. Any thoughts on what might be a better move?
        
        roseh: if you don’t have long term need cloud compute is cheaper
        
        kenta: I think Amazon is not the cheapest source for straight compute power. DigitalOcean might be better?
        
        medesaro: Also what about uploading a 200 GB file to the Internet? It's frustrating to see this operation taking so long.
        
        cjq: how many 200GB files are we talking about
        
        medesaro: I have a box of tapes, and maybe 20 tapes.
        
        toomas: How long is the tape?
        
        medesaro: Each is like 20 hours? There's this new project called VHS-Decode. Problem is you end up dealing with more data than I've ever tried to deal with before. Looks like it's about $1000 on eBay for a used thread-ripper.
        
        huydai: buying a personal machine won't be too bad?
        
        medesaro: the machine i'm running is like 12 cores, but it's more of the virtual core doubling, it's a xenon workstation processor but it's from about 2009, so i suspect that's part of the problem. it was top of the line then!
        
        jmvidal: prices today aren't too bad, like a few hundred [for a motherboard and processor]
        
        medesaro: usually my work doesn't involve such large files...
        
        agrebe: 400 hours of time on your desktop that's 6 cores? aws is 5 cents per core-hour, so we're talking about $200ish. less than a computer, but you get to keep the computer. you can get cheaper if you only need one core and less memory, but i don't imagine that's what you want.
        
        medesaro: or i could rewrite it in c++...
        
        colclark: rust.
        
        medesaro: It's massive amount of FFT. Google VHS-Decode, and you can see the code. It's open-source. I made a few custom changes but it's basically stock.
        
        huydai: microsoft wants you to bing, not google.
        
        kenta: If you want to piss everyone off, you should look up "Google" on Bing.
        
        roseh: in the address bar, search “google” then click google, search “duckduckgo” click duckduckgo, search “yandex” 
        
        [discussion of DuckDuckGo vs. Bing]
        
        jmvidal: "google" is the third most popular search on bing.


The meeting was adjourned at 20:03.

        Minutes taken and submitted by cjq.
