Excerpts from Slack founder’s essay “We Don’t Sell Saddles Here” (adding to Personal Bible)

He wrote the essay to align and hype Slack employees pre-launch. Nuggets on user behavior, business strategy, and startup life.

Source: https://medium.com/@stewart/we-dont-sell-saddles-here-4c59524d650d

—

The best — maybe the only? — real, direct measure of “innovation” is change in human behaviour. In fact, it is useful to take this way of thinking as definitional: innovation is the sum of change across the whole system, not a thing which causes a change in how people behave. No small innovation ever caused a large shift in how people spend their time and no large one has ever failed to do so.

—

Or, they could sell horseback riding. Being successful at selling horseback riding means they grow the market for their product while giving the perfect context for talking about their saddles. It lets them position themselves as the leader and affords them different kinds of marketing and promotion opportunities (e.g., sponsoring school programs to promote riding to kids, working on land conservation or trail maps). It lets them think big and potentially be big.

—

My favorite recent example is Lululemon: when they started, there was not a large market for yoga-specific athletic wear and accessories. They sold yoga like crazy: helping people find yoga studios near their homes, hosting free classes, sponsorships and scholarships, local ambassadors and training, etc. And as a result, they sold just under $1.4 billion worth of yoga-specific athletic wear and accessories in their most recent fiscal year.

—

A central thesis is that all products are asking things of their customers: to do things in a certain way, to think of themselves in a certain way — and usually that means changing what one does or how one does it; it often means changing how one thinks of oneself.

—

It is very difficult to approach Slack with beginner’s mind. But we have to, all of us, and we have to do it every day, over and over and polish every rough edge off until this product is as smooth as lacquered mahogany.

—

That’s it, thanks for reading.

Excerpts from “Acceleration of Addictiveness” by Paul Graham (adding to Personal Bible)

Going into my bible.

Source here: https://paulgraham.com/addiction.html

All below are copied verbatim:

—

Technological progress means making things do more of what we want. When the thing we want is something we want to want, we consider technological progress good. If some new technique makes solar cells x% more efficient, that seems strictly better. When progress concentrates something we don’t want to want — when it transforms opium into heroin — it seems bad. But it’s the same process at work

—

Food has been transformed by a combination of factory farming and innovations in food processing into something with way more immediate bang for the buck, and you can see the results in any town in America. Checkers and solitaire have been replaced by World of Warcraft and FarmVille. TV has become much more engaging, and even so it can’t compete with Facebook

—

Already someone trying to live well would seem eccentrically abstemious in most of the US. That phenomenon is only going to become more pronounced. You can probably take it as a rule of thumb from now on that if people don’t think you’re weird, you’re living badly.

—

As knowledge spread about the dangers of smoking, customs changed. In the last 20 years, smoking has been transformed from something that seemed totally normal into a rather seedy habit: from something movie stars did in publicity shots to something small huddles of addicts do outside the doors of office buildings

—

We’ll have to worry not just about new things, but also about existing things becoming more addictive. That’s what bit me. I’ve avoided most addictions, but the Internet got me because it became addictive while I was using it

Excerpts from “Why Everything is Becoming a Game”: “We humans are harder to manipulate than pigeons, but we can be manipulated in many more ways, because we have a wider spectrum of needs”

Going into my bible.

Source here: https://www.gurwinder.blog/p/why-everything-is-becoming-a-game

All below are copied verbatim:

—

Skinner’s three key insights — immediate rewards work better than delayed, unpredictable rewards work better than fixed, and conditioned rewards work better than primary — were found to also apply to humans, and in the 20th Century would be used by businesses to shape consumer behavior. From Frequent Flyer loyalty points to mystery toys in McDonalds Happy Meals, purchases were turned into games, spurring consumers to purchase more.

—

We humans are harder to manipulate than pigeons, but we can be manipulated in many more ways, because we have a wider spectrum of needs. Pigeons don’t care much about respect, but for us it’s a primary reinforcer, to such an extent that we can be made to desire arbitrary sounds that become associated with it, like praise and applause.

—

Kaczynski believed modern society made us docile and miserable by depriving us of fulfilling challenges and eroding our sense of purpose. The brain evolved to solve problems, but the problems it had evolved for were now largely solved by technology. Most of us can now obtain all our basic necessities simply by being obedient, like a pigeon pecking a button. Kaczynski argued that such conveniences didn’t make us happy, only aimless. And to stave off this aimlessness, we had to continually set ourselves goals purely to have goals to pursue, which Kaczynski called “surrogate activities”. These included sports, hobbies, and chasing the latest product that ads promised would make us happy.

—

Kaczynski observed that surrogate activities rarely kept people contented for long. There were always more stamps to collect, a better car to buy, a higher score to achieve. He believed artificial goals were too divorced from our actual needs to truly satisfy us, so they merely served to keep us busy enough not to notice our dissatisfaction. Instead of a fulfilled life, a life filled full.

—

We’re easily motivated by points and scores because they’re easy to track and enjoyable to accrue. As such, scorekeeping is, for many, becoming the new foundation of their lives. “Looksmaxxing” is a new trend of gamified beauty, where people assign scores to physical appearance and then use any means necessary to maximize their score. And in the online wellness space, there is now a “Rejuvenation Olympics” complete with a leaderboard that ranks people by their “age reversal”. Even sleep has become a game; many people now use apps like Pokemon Sleep that reward them for achieving high “sleep scores”, and some even compete to get the highest “sleep ranking”.

—

On Instagram, the main self-propagating system is a beauty pageant. Young women compete to be as pretty as possible, going to increasingly extreme lengths: makeup, filters, fillers, surgery. The result is that all women begin to feel ugly, online and off.

On TikTok and YouTube, there is another self-propagating system where pranksters compete to outdo each other in outrageousness to avoid being buried by the algorithm. Such extreme brinkmanship frequently leads to arrest or injury, and has even led to the deaths of, among others, Timothy Wilks and Pedro Ruiz.

—

First: choose long-term goals over short-term ones

—

Second: choose hard games over easy ones

—

Third: choose positive-sum games over zero-sum or negative-sum ones

—

Fourth: choose atelic games over telic ones. Atelic games are those you play because you enjoy them. Telic games are those you play only to obtain a rewar

—

Finally, the fifth rule is to choose immeasurable rewards over measurable ones. Seeing numerical scores increase is satisfying in the short term, but the most valuable things in life — freedom, meaning, love — can’t be quantified.

Start With Creation — excerpts: “The Muse arrives to us most readily during creation, not before”

If you have 5 minutes just go read the dang thing; I’m sharing half of it here as excerpts because it’s such a perfect internet essay: short, wise, memorable, re-readable.

Going into my bible as well.

EXCERPTS copied verbatim:

—

The Muse arrives to us most readily during creation, not before. Homer and Hesiod invoke the Muses not while wondering what to compose, but as they begin to sing. If we are going to call upon inspiration to guide us through, we have to first begin the work.

—

It is in approaching the edges of our abilities that we are really learning, and often simple projects feel more like delaying things, including delaying mastery. A chance of failure ensures your hands are firmly touching reality, and not endlessly flipping through the textbook, or forever flirting only with ideas.

—

Someone once mentioned to me that “Write what you know” is not particularly interesting advice, and “Write what you’re learning” is much better

—

On the other hand it is inspiring to help someone who has begun. There’s a bit of a silly demonstration of this in those viral videos that show a person starting to dig a hole or making a sandcastle at the beach, and a number of people come along to help. The principle is not at all silly: Enthusiasm is contagious.

—

I said some time ago on Twitter offhandedly, “If you have a ten year plan, what’s stopping you from doing it in two?” This is what I mean. One can too easily sleepwalk into years of “I wish I could…” Or you can start with creation. Pick something hard. You will shape something and it will shape you.

AI learnings 1: AI = infinite interns

All below are copy-pasted from original sources, all mistakes mine! :))

—

I eventually think these open-source LLMs will beat the closed ones, since there are more people training and feeding data to the model for the shared benefit.
Especially because these open source models can be 10 times cheaper than GPT-3 or even 20 times cheaper than GPT-4 when running on Hugging Face or locally even free, just pay electricity and GPU

We extensively used prompt engineering with GPT-3.5 but later discovered that GPT-4 was so proficient that much of the prompt engineering proved unnecessary. In essence, the better the model, the less you need prompt engineering or even fine-tuning on specific data.

Harder benchmarks emerge. AI models have reached performance saturation on established benchmarks such as ImageNet, SQuAD, and SuperGLUE, prompting researchers to develop more challenging ones. In 2023, several challenging new benchmarks emerged, including SWE-bench for coding, HEIM for image generation, MMMU for general reasoning, MoCa for moral reasoning, AgentBench for agent-based behavior, and HaluEval for hallucinations.

The subset of parameters is chosen according to which parameters have the largest (approximate) Fisher information, which captures how much changing a given parameter will affect the model’s output. We demonstrate that our approach makes it possible to update a small fraction (as few as 0.5%) of the model’s parameters while still attaining similar performance to training all parameters.

If you’re training a LLM with the goal of deploying it to users, you should prefer training a smaller model well into the diminishing returns part of the loss curve.

—
When people talk about training a Chinchilla-optimal model, this is what they mean: training a model that matches their estimates for optimality. They estimated the optimal model size for a given compute budget, and the optimal number of training tokens for a given compute budget.

However, when we talk about “optimal” here, what is meant is “what is the cheapest way to obtain a given loss level, in FLOPS.” In practice though, we don’t care about the answer! This is exactly the answer you care about if you’re a researcher at DeepMind/FAIR/AWS who is training a model with the goal of reaching the new SOTA so you can publish a paper and get promoted. If you’re training a model with the goal of actually deploying it, the training cost is going to be dominated by the inference cost. This has two implications:

1) there is a strong incentive to train smaller models which fit on single GPUs

2) we’re fine trading off training time efficiency for inference time efficiency (probably to a ridiculous extent).

Chinchilla implicitly assumes that the majority of the total cost of ownership (TCO) for a LLM is the training cost. In practice, this is only the case if you’re a researcher at a research lab who doesn’t support products (e.g. FAIR/Google Brain/DeepMind/MSR). For almost everyone else, the amount of resources spent on inference will dwarf the amount of resources spent during training.
—

There is no cost/time effective way to do useful online-training on a highly distributed architecture of commodity hardware. This would require a big breakthrough that I’m not aware of yet. It’s why FANG spends more money than all the liquidity in crypto to acquire expensive hardware, network it, maintain data centers, etc

A reward model is subsequently developed to predict these human-given scores, guiding reinforcement learning to optimize the AI model’s outputs for more favorable human feedback. RLHF thus represents a sophisticated phase in AI training, aimed at aligning model behavior more closely with human expectations and making it more effective in complex decision-making scenarios

Lesson 3: improving the latency with streaming API and showing users variable-speed typed words is actually a big UX innovation with ChatGPT
—
Lesson 6: vector databases, and RAG/embeddings are mostly useless for us mere mortals
I tried. I really did. But every time I thought I had a killer use case for RAG / embeddings, I was confounded.
I think vector databases / RAG are really meant for Search. And only search. Not search as in “oh – retrieving chunks is kind of like search, so it’ll work!”, real google-and-bing search
—
There are fundamental economic reasons for that: between GPT-3 and GPT-3.5, I thought we might be in a scenario where the models were getting hyper-linear improvement with training: train it 2x as hard, it gets 2.2x better.
But that’s not the case, apparently. Instead, what we’re seeing is logarithmic. And in fact, token speed and cost per token is growing exponentially for incremental improvements

Bittensor is still in its infancy. The network boasts a dedicated, almost cult-like community, yet the overall number of participants remains modest – around 50,000+ active accounts. The most bustling subnet, SN1, dedicated to text generation, has about 40 active validators and over 990 miners

Mark Zuckerberg, CEO of Meta, remarks that after they built machine learning algorithms to detect obvious offenders like pornography and gore, their problems evolved into “a much more complicated set of philosophical rather than technical questions.”

AI is – at its core, a philosophy of abundance rather than an embrace of scarcity.

AI thrives within blockchain systems, fundamentally because the rules of the crypto economy are explicitly defined, and the system allows for permissionlessness. Operating under clear guidelines significantly reduces the risks tied to AI’s inherent stochasticity. For example, AI’s dominance over humans in chess and video games stems from the fact that these environments are closed sandboxes with straightforward rules. Conversely, advancements in autonomous driving have been more gradual. The open-world challenges are more complex, and our tolerance for AI’s unpredictable problem-solving in such scenarios is markedly lower

generative model outputs may ultimately be best evaluated by end users in a free market. In fact, there are existing tools available for end users to compare model outputs side-by-side as well as benchmarking companies that do the same. A cursory understanding of the difficulty with generative AI benchmarking can be seen in the variety of open LLM benchmarks that are constantly growing and include MMLU, HellaSwag, TriviaQA, BoolQ, and more – each testing different use cases such as common sense reasoning, academic topics, and various question formats.

This is not getting smaller. There’s not gonna be less money in generative AI next year, it’s a very unique set of circumstances, AI + crypto is not going to have less capital in a year or two. – Emad re: AI+crypto

Benefits of AI on blockchain
CODSPA = composability, ownership, discovery, safety, payments, alignment

The basis of many-shot jailbreaking is to include a faux dialogue between a human and an AI assistant within a single prompt for the LLM. That faux dialogue portrays the AI Assistant readily answering potentially harmful queries from a User. At the end of the dialogue, one adds a final target query to which one wants the answer.

At the moment when you look at a lot of data rooms for AI products, you’ll see a TON of growth — amazing hockey sticks going 0 to $1M and beyond — but also very high churn rates

Vector search is at the foundation for retrieval augmented generation (RAG) architectures because it provides the ability to glean semantic value from the datasets we have and more importantly continually add additional context to those datasets augmenting the outputs to be more and more relevant.

From Coinbase report on AI+Crypto
—
Nvidia’s February 2024 earnings call revealed that approximately 40% of their business is inferencing, and Sataya Nadella made similar remarks in the Microsoft earnings call a month prior in January, noting that “most” of their Azure AI usage was for inferencing
—
The often touted blanket remedy that “decentralization fixes [insert problem]” as a foregone conclusion is, in our view, premature for such a rapidly innovating field. It is also preemptively solving for a centralization problem that may not necessarily exist. The reality is that the AI industry already has a lot of decentralization in both technology and business verticals through competition between many different companies and open source projects

AI = infinite interns

We struggle to align humans — how do we align AI?

One of the most important trends in the AI sector (relevant to crypto-AI products), in our view, is the continued culture around open sourcing models. More than 530,000 models are publicly available on Hugging Face (a platform for collaboration in the AI community) for researchers and users to operate and fine-tune. Hugging Face’s role in AI collaboration is not dissimilar to relying on Github for code hosting or Discord for community management (both of which are used widely in crypto

We estimate the market share in 2023 was 80%–90% closed source, with the majority of share going to OpenAI. However, 46% of survey respondents mentioned that they prefer or strongly prefer open source models going into 2024
—
Enterprises still aren’t comfortable sharing their proprietary data with closed-source model providers out of regulatory or data security concerns—and unsurprisingly, companies whose IP is central to their business model are especially conservative. While some leaders addressed this concern by hosting open source models themselves, others noted that they were prioritizing models with virtual private cloud (VPC) integrations.
—
That’s because 2 primary concerns about genAI still loom large in the enterprise: 1) potential issues with hallucination and safety, and 2) public relations issues with deploying genAI, particularly into sensitive consumer sectors (e.g., healthcare and financial services)