“Mark Zuckerberg in Warhammer Total War portrait smirk” c/o Lexica
It feels like one of those timeless patterns of history where a ruler in any domain — whether an Emperor, a Founder CEO, or even a teacher in a classroom — initially serves his people, but when that ruler acquires too much power over a long period of time, he starts to believe the people serve him. I’m sure there’s a wise Confucius proverb describing precisely this…
Why do I need to log in to Reddit to read comments? Why can’t I fix a spelling error on Twitter? Why can’t I find the names of the band members on Spotify? Why is the whole first page of Google search results sometimes filled with paid advertising? Why does TikTok send all my private data to China?
–
It’s obvious that these companies didn’t do focus groups or market research before making these decisions. Or if they did, they must have ignored what they learned.
–
I’ve focused here on Facebook, but there’s a larger lesson here. Web platforms don’t fail because of the competition. They don’t self-destruct because they are weak. The collapse comes because they are strong. They lose the thread because of their dominance and power, which gives their leaders the mindset of authoritarian rulers.
–
The solution is simple: *Serve the users, instead of manipulating them*
I’m far from an AI expert, just an interested student who gets the tingly feels every time I use Stable Diffusion or see output from ChatGPT.
Snippets (copied verbatim):
–
The chasm between academia and industry in large scale AI work is potentially beyond repair: almost 0% of work is done in academia.
–
Finding faster matrix multiplication algorithms, a seemingly simple and well-studied problem, has been stale for decades. DeepMind’s approach not only helps speed up research in the field, but also boosts matrix multiplication based technology, that is AI, imaging, and essentially everything happening on our phones.
–
The authors argue that the ensuing reproducibility failures in ML-based science are systemic: they study 20 reviews across 17 science fields examining errors in ML-based science and find that data leakage errors happened in every one of the 329 papers the reviews span
–
many LLM capabilities emerge unpredictably when models reach a critical size. These acquired capabilities are exciting, but the emergence phenomenon makes evaluating model safety more difficult.
–
Alternatively, deploying LLMs on real-world tasks at larger scales is more uncertain as unsafe and undesirable abilities can emerge. Alongside the brittle nature of ML models, this is another feature practitioners will need to account for.
–
Landmark models from OpenAI and DeepMind have been implemented/cloned/improved by the open source community much faster than we’d have expected.
–
Compared to US AI research, Chinese papers focus more on surveillance related-tasks. These include autonomy, object detection, tracking, scene understanding, action and speaker recognition.
–
NVIDIA’s chips are the most popular in AI research papers…and by a massive margin
–
“We think the most benefits will go to whoever has the biggest computer” – Greg Brockman, OpenAI CTO
–
As such, the AI could reliably remove 36.4% of normal chest X-rays from a primary health care population data set with a minimal number of false negatives, leading to effectively no compromise on patient safety and a potential significant reduction of workload.
–
The US leads by the number of AI unicorns, followed by China & the UK; The US has created 292 AI unicorns, with the combined enterprise value of $4.6T.
–
The compute requirements for large-scale AI experiments has increased >300,000x in the last decade. Over the same period, the % of these projects run by academics has plummeted from ～60% to almost 0%. If the AI community is to continue scaling models, this chasm of “have” and “have nots” creates significant challenges for AI safety, pursuing diverse ideas, talent concentration, and more.
–
Decentralized research projects are gaining members, funding and momentum. They are succeeding at ambitious large-scale model and data projects that were previously thought to be only possible in large centralised technology companies – most visibly demonstrated by the public release of Stable Diffusion.
Great overall podcast, because Peterson is a world class explainer and to explain well he needs to understand well and to understand well he really digs and pokes thoroughly. Not saying I believe all of it, because China number one and all that.
Reasons why it *could* be a lab leak (but definitely not saying it is, y’know), my paraphrased notes:
—
-there was a lab near the outbreak researching exactly this kind of virus
-the virus was atypical in its ability to spread between humans (transmissibility)
-they identified a “furin cleavage site” (an added bit of DNA code) in the virus DNA
-there’s still a lack of finding the animal transmission vector / specimen(s)
-the multiple attempts to cover up early findings (in both China and the US) that even hinted at a potential lab leak / man-made virus
-there was a red herring of an identified pangolin virus whose DNA sequence was later found to be too different, and was not found near the same area
-those scientists and administrators in charge in both the US and China had grant proposals and research projects on precisely this (furin cleavage, bat viruses, gain of function)
-there were prior bio safety incidents at that very lab, on which the top Chinese leadership were consulted
-this new virus differs greatly from others like it, which were bat viruses, but not particularly lethal, and mostly intestinal
-in the case of SARS, the transmission chain very clear, and the animal vector and index cases were eventually found; none of that’s happened here
-although there was a heavy concentration of cases near the suspected origination wet market, it’s a bit of drawing the bullseye after taking the shot; only those who self-identified as being near that market were diagnosed with it — if you weren’t near the market, even if you had the same symptoms, you were diagnosed with something else (like the flu?)
A friend and I are starting a crypto podcast because we’re middle aged crypto nerds who want to hear ourselves talk.
I’ll be posting the links here, and often. We’ve already recorded 4 episodes and will publish them soon, and will also invite guests in the future
There’ll be brief takes on the state of the market and recent crypto news, but the podcast focus will be crypto-related media (like blogs, tweets, and podcasts) that we recommend and why. There’s just an endless torrent of good content, and we want to get wet and then share the water drops (great metaphor right)
Here are some examples that I’ll probably mention in future episodes:
—
Reality check on state of crypto today and what causes the bubbles / cycles:
Imagine a world where the most active investors in traditional finance are Nasdaq Ventures and the NYSE, and the financial information on those listed securities is opaque. That is our reality in the crypto sector. This is what we have created
—
Approximately 9 minutes into this episode of All-In, Chamath gives the best explanation of what’s happening with the SBF / FTX fraud circus and mainstream media’s role in it:
Perhaps my favorite bitcoin analyst, David has rare perspective from decades in tradfi and a unique lens to look at bitcoin price action (namely, focusing on whales):
That said, and as I intimated above, it is always darkest before dawn and investor sentiment is as dark now as at any point over my thirty years of investing. I suspect a lot of readers did not experience firsthand how frightening the dot-com bubble and GFC were, but I did, so I can assure you that the FUD then was no less compelling than the FUD right now
—
Philosophical reflection on crypto market cycles and why we’re here
Free will perhaps exists on micro levels – individual and local – but on a broader scale, the human condition seems to be solidified in its ways and patterns. Markets are a great example where this constantly plays out – especially in the relatively nascent, less-regulated, more free-market crypto space where multiple cycles have occurred, all culminating and correcting in eerily similar ways.
—
Thorough explanation of the short and long-term debt cycles as popularized by Ray Dalio (through a bitcoiner’s lens)
In a free-market capitalist economic system, the most important pricing mechanism is that of money. When there is a monopolist institution setting the price of money, the market is inherently not “free.” There is nothing free about reducing the price of money whenever there is an economic downturn, including the most recent injections of hundreds of billions and now trillions of dollars into financial markets whenever a major liquidation of malinvestment occurs.
—
One of the more thorough newsletters covering crypto news with brief but thoughtful takes:
First, in your teens or 20s, to take it all in. See it all, do it all, and learn. Get involved. Stay up all night talking with strangers, everywhere. Kiss and fall and promise to them all. Make lots of mistakes.
Cross the world the first time to fall in love.
A beautiful concept, and something I personally experience each time I travel.
Now, in my late 30s, I put myself somewhere between Derek’s description of the second and third times. Perhaps closer to the third time, it just feels heavier.
And that fourth time — what a bittersweet sadness 🥹
