plz 8
Inside
Alright we really have to talk about Bo Burnham’s Inside. Well, I really have to, at least. Inside is a 90-minute… comedy special? Movie? Something? It came out on Netflix at the end of May. The premise is that Bo made the entire thing himself during the pandemic: he wrote it, shot it, and edited it inside a one-room building in his backyard. It’s like a mother in law house, a tiny studio apartment with a bed and a small kitchen and living area. The entire special (I’ll call it a special, it’s a special) takes place in what is essentially one room. There is of course no audience but there is music, dynamic lighting, different scenes and characters, songs in Bo’s typical comedy style, skits, and… other stuff. We’ll get to the other stuff.
But it’s all Bo. If there’s a camera recording it’s because he set it up and plugged it in and hit record and then walked to the other side of the room and sat down to play a song. It’s an essential component: the themes of isolation and loneliness are heavy in the music but also in the (frequent) behind-the-scenes style shots that show Bo literally making the special. Between the songs — some of which feel highly-produced, almost cinematic — are shots of Bo adjusting the lighting, checking the camera shot, or testing the microphone. The process of making the special is treated as a first-class sibling of the songs and comedy.
Bo’s aloneness appears in satisfyingly subtle ways as well. There are two ways to zoom a camera: optical zoom and digital zoom. With optical zoom, the physical lens in the camera moves toward or away from the camera’s sensor, magnifying or shrinking the image that is captured. Optical zoom doesn’t affect the image quality but typically requires a manual touch since the zoom is part of the recording itself. Digital zoom is really just a software trick: taking a captured image and cropping it, then scaling it up to make it appear zoomed in. That’s nice because it can be done during editing (and changed, unlike optical zoom). The problem with digital zoom is that you only have the pixels that you started with so if you zoom too much you’ll start to see the ‘texture’ of the pixels because the image you’re seeing is essentially lower resolution.
There are a lot of zooms in Inside and they’re all digital. The most common type is a very slight zoom in or out from a wide-angle shot, so slow you’ll miss it if you’re not looking for it. If you’re filming a comedy special alone in a room then a gentle zoom over the course of a ten or twenty second shot is a nice way to provide some motion (a totally still shot can feel off-putting, especially when most of the things in the frame aren’t moving either). The second shot of the special is something else, though. Bo is sitting in a mostly dark room with the camera zoomed far in on his head and shoulders. The darkness makes the quality loss effect of the digital zoom even more pronounced and the pixelation makes the wall behind him appear to ripple.
The quality loss from the zoom in this second shot of the special is reminiscent of early YouTube videos, like the 240p ones that Bo uploaded to YouTube 15 years ago. Videos he wrote, shot, and edited… in his room. Those videos are referenced in the special[0] and one even plays on screen briefly. The camera, though still working, has been pushed past the point where it is designed to function well. And as you watch the special you see that the same is true for Bo.
This is a good example of what I think makes Inside so excellent. We’re only on the second shot but we can already reach themes that run throughout the special. I think a litmus test for great art is that it’s not only resilient to analysis but that the deeper you go into it, the more you find. With Bo exerting total control over the music, cinematography, and editing he left a lot for the viewer to discover.
I was originally going to write this super long scene-by-scene analysis of the whole special but when I tried that I think the format made me more critical than is maybe helpful or interesting. So I’m not going to go over most of the content, you can enjoy that for yourself. I do think some elements of the structure and general meta-level deserves specific note, though.
Although there are plenty of jokes in Inside, the default tone is something closer to malaise or straight up depression. That’s a setup that shouldn’t play well for a comedy special, yet it does. Bo capitalizes on a set of ubiquitous 2020-era experiences to apply conventional mass-appeal comedy tactics (‘What’s the deal with airline food’, etc.) to some very dark emotions and thoughts. Comedians can kind of do this normally (there are widely shared negative experiences like death or divorce) but the pandemic and social justice movements of 2020 unlocked a ripe new arena for negative comedy. With live comedy still recovering from the pandemic, Inside is really the first in-depth exploration of these events in a comedic format.
The audience, despite being conspicuously absent, is probably the closest thing to a supporting cast member. The finale to Bo’s last special, Make Happy, was a confrontation of his conflicting desires for appreciation and freedom from judgement[1]. In Inside he seems tormented by the implied presence of the audience — despite his solitude he clearly feels like he’s being watched and tries in several ways to confront the viewer: addressing the camera, self-consciously adding in behind-the-scenes footage, and including several prolonged zooms into a reflection of the camera lens itself.
Linked to Bo’s emotional state is a recurring question about what is real. Born in 1990 and with a career that began in the first age of viral YouTube stars, Bo is uniquely positioned to reflect on the ways the internet (and particularly media on the internet) dictate what reality is. In Inside that takes the form of deadpan interludes, absurdist renditions of YouTube and Twitch content, and recurring cues that remind the viewer: this is all a performance. Bo will deliver an emotional, apparently unscripted scene only to burst into a song. A Bob Dylan-esque folk ballad accompanied by flickering fireside ambience is preceded by a close-up of the light fixture supplying the artificial atmosphere. A self-described ‘test’ take of a song is slowly overlaid with accompaniment from a longer-haired Bo, clearly recorded in the past. An hour in you begin to wonder: is any of this really candid? Are these apparently behind-the-scenes shots staged as much as the songs and skits are? The performance and documentary aspects of Inside become indistinguishable. There are multiple scenes where the end of song smash-cuts to Bo, in the dark, watching the video of the song that just played on his laptop. You’re rarely allowed to forget the nature of the special itself: Bo is alone and increasingly consumed by creating the thing that is on your screen.
GitHub Copilot
Editor’s Note: This is a couple months delayed but it’s exactly the sort of thing I originally intended to write about here so better late than never.
One problem that a lot of companies have is that software is really valuable but creating software is costly. You have to hire programmers and wait for them to write your software. The programmers are expensive and sometimes they make mistakes. If you could help your programmers write software faster (or more correctly) that could be a big help to your business! And it’s not like every software project is something totally new, people have written lots of software already and a lot of that code is on the internet. Is there some way you could take advantage of that to make it easier or cheaper to create software?
Today, we are launching a technical preview of GitHub Copilot, a new AI pair programmer that helps you write better code. GitHub Copilot draws context from the code you’re working on, suggesting whole lines or entire functions. It helps you quickly discover alternative ways to solve problems, write tests, and explore new APIs without having to tediously tailor a search for answers on the internet. As you type, it adapts to the way you write code—to help you complete your work faster.
Developed in collaboration with OpenAI, GitHub Copilot is powered by OpenAI Codex, a new AI system created by OpenAI. OpenAI Codex has broad knowledge of how people use code and is significantly more capable than GPT-3 in code generation, in part, because it was trained on a data set that includes a much larger concentration of public source code. GitHub Copilot works with a broad set of frameworks and languages, but this technical preview works especially well for Python, JavaScript, TypeScript, Ruby and Go.
Researchers published a scholarly paper examining the security implications of GitHub Copilot, an excellent AI system now being used for code completion in Visual Studio Code. In various scenarios, some 40 percent of tested projects were found to include security vulnerabilities.
I’m just having a little fun. The jury is still out on whether Copilot is good or bad or usable or a tremendous legal and technical liability. Wait, what’s going on here?
Okay so GitHub (owned by Microsoft, remember) trained a model on all of the publicly available code on its site. The result is something kind of like the auto-suggest feature on your phone, though more flexible. It looks at the code you’re writing (or even just the comments) and tries to guess what code should come next. The demo site is very polished, you should take a look.
People have been using this for a while now and I’ve seen three basic reactions. The first is ‘huh this works pretty well’ (e.g. here). Our collective ability to build AI for these types of predictive tasks is improving and I expect we’ll see this type of thing expand to many more domains soon. Copilot is a conceptual extension of things like Gmail’s Smart Compose.
The second is ‘heh, the AI will do funny or dumb things in the right conditions’. The pitch with Copilot is like ‘we instilled a machine with the collective wisdom of a trillion lines of code, now it writes beautiful original software’ but sometimes the reality is that it regurgitates entire blocks of the Quake source code, complete with profane comments.
That leads into the third reaction, which is ‘wait is this a massive legal liability?’. Ostensibly, Copilot is writing ‘original’ code in approximately the same sense that a human does. It has been trained on public GitHub repositories with varying licenses, but the restrictions of those licenses shouldn’t apply to the code produced by Copilot since the code isn’t directly derived from the training data. Sometimes this seems to be the case, but there are many documented instances (see Quake, above) where Copilot spits out giant chunks of code verbatim from identifiable sources. This poses enough of a legal risk (what if your programmers unknowingly copy licensed code?) that companies will likely be hesitant to allow employees to use a system like Copilot on the job until there are stronger legal guarantees about Copilot’s output.
There were also some jokes of the form ‘the programmers never thought automation would come for them, what delicious irony’ which is funny but probably not predictive. I think Patrick has the right idea: every innovation that has in theory reduced the need for programmers by making them more efficient (compilers, linters, cloud services, etc.) has actually done the opposite.
If you want to dig further into the copyright questions, Fossa and The Register are good places to start. If you just want a deeper dive into Copilot, check out fast.ai.
Bookmarks
What’s going on here, with this human? IRS Poses as Bitcoin Trader ‘Mr Coins’ in $180,000 Sting. Cool interactive site about naval architecture. Quantifying the effectiveness of DIY air purifiers. Advanced situational awareness.
[0] Bo’s reflection on and dissatisfaction with his past work is a major theme (“Am I right back where I started fourteen years ago?”). The shot before ‘Problematic’ where he watches a projection of his old YouTube content (in apparent disgust) parallels the final shot of the special where he smiles while watching the video of himself trying to break back into the room.
[1] A gut-wrenching easter egg: the outro to Make Happy was shot in the same room as Inside. After an emotional finale on stage Bo appears in the room, plays a short song about happiness on the piano, and then walks out the door into the sunshine. He greets his wife and dog and appears to leave the pain of that special behind in the room. The opening shot of Inside is from the same angle as Bo slowly opens the door and returns to this emotional (and now literal) prison.