Builds VR training and roleplaying games. Cyberpunk and futurism nerd. https://xitter.cf/t.co/gZ2yVKh7dz and https://xitter.cf/t.co/AwZFZEGbri
Stockholm, Sweden
Joined June 2008
- Tweets3.6K
- Following977
- Followers258
- Likes1.6K
ripperdoc retweeted
Hur kan man lägga 5,5 miljarder för ett digitalt journalsystem som inte ens tar med sig de tidigare journaler? Vad lever vi i för bisarr värld?
ripperdoc retweeted
Today I'm open sourcing @timelinize, an ambitious project that organizes your life's data from your online accounts, photo library, phone, etc, onto a single unified timeline on your own computer. It will benefit from the community's help. 🧵 github.com/timelinize/timeli…
ripperdoc retweeted
This is so true. It's hard enough building regular software and then finding that end users instantly break it in weird ways you didn't expect - the scope for ways they can break anything with an LLM involved is effectively unbounded
ripperdoc retweeted
Small victory but 2 years later no one says “web2” any more… because “web3” has mostly disappeared. It was a term fueled by select VCs invested heavily in crypto companies.
I just feel sorry for devs who fell for it and believed it’s somehow superior to normal engineering.
The new field of AIstrology?
Optional chaining operator is such a developer experience improvement in JavaScript (and other languages that have it). Can’t live without it. developer.mozilla.org/en-US/…
Completely agree!
A viral paper "Language Model Represents Space and Time" recently claims that LLMs learn "world models". As much as I like @tegmark's works, I disagree with their definition of world model.
World model is a core concept in AI agent and decision making. It is our mental simulation of how the world works given interventions (or lack thereof).
A world model captures causality and intuitive physics, telling the agent what is likely and what is impossible. It can and should be used for counterfactual reasoning, i.e. "what ifs": what would happen if I knock over a cup of water? Where would I have been if I had not taken that bus?
Yann LeCun @ylecun says it well in his position paper (openreview.net/pdf?id=BZ5a1r…). I quote:
"Using such world models, animals can learn new skills with very few trials. They can predict the consequences of their actions, they can reason, plan, explore, and imagine new solutions to problems. Importantly, they can also avoid making dangerous mistakes when facing an unknown situation."
The first use of the term World Model in deep policy learning is attributed to @hardmaru & @SchmidhuberAI: worldmodels.github.io/. In their seminal paper, an agent masters shooting skills in the popular game Doom (demo below) by learning in imagination, using an internal world model as a "physics simulator".
To put in a simple Python math formula, world model learns a function F(s[0:t-1], a) -> s[t:], which takes as input the observed past and current action, and outputs plausible future states.
Now the definition of World Model in Tegmark's paper seems to be about predicting GPS coordinates and time eras. I see this as just a classification task with no causal learning and simulation going on. You cannot make meaningful interventions against that model, nor can you optimize any decision making in a closed feedback loop.
As for the "space & time neurons", I think they are most similar to the "sentiment neuron" that OpenAI published in 2017: openai.com/research/unsuperv…. Predicting GPS is conceptually no different from predicting sentiment in my opinion. I don't think their experimental results are wrong - just that their conclusion is on shaky grounds.
I welcome any debate! Paper link: arxiv.org/abs/2310.02207
ripperdoc retweeted
The adoption of cellphones by Keralan fishermen is, I believe, the most stunning example of the contribution of information technology to market performance.
Take a look at this graph for background: in three different regions of Kerala, phones were adopted at different times.
Kozhikode got cellphones before Kannur, who got them before Kasaragod. Adoption by fishermen was fast when phones were finally introduced:
Now look at what happened to the price of fish after phones entered the scene:
The dispersion in prices virtually disappeared! The author of the study wrote about this that:
"Before any region had mobile phones, the degree of price dispersion across markets within a region on any given day is high, and there are many cases where the price is zero (i.e., waste). However, within a few weeks of mobile phones being introduced in Region I, there is a sharp and striking reduction in price dispersion. Prices across markets in the region rarely differ by more than a few rupees per kilogram on any day, compared to cases of as much as 10 Rs/kg prior to the introduction of mobile phones. In addition, the prices in the various markets rise and fall together and the week-to-week variability within each market is much smaller, since catchment zone-specific quantity shocks are now spread across markets via arbitrage. Further, there are no cases of waste in this region after phones are introduced.
"By contrast, price behavior in regions II and III appears largely unchanged after phones are introduced in region I. However, after mobile phones are introduced in region II, prices again become much less dispersed across markets on any given day, less variable within markets over time, and waste is ultimately eliminated, whereas region III again remains unchanged. Finally, the same pattern holds once region III adds phones."
Efficient information transmission enables efficient markets. It's amazing what technology can do.
Read the study here: jstor.org/stable/25098864
I usually try to stay away from bleeding edge in JS ecosystem, but this seems way too compelling to not try out.
Introducing Bun v1.0 - a fast, all-in-one toolkit for running, building, and testing JavaScript and TypeScript
bun.sh/blog/bun-v1.0
ripperdoc retweeted
What Wikipedia did: we stood strong for our principles and fought to the Supreme Court of Turkey and won. This is what it means to treat freedom of expression as a principle rather than a slogan.
It was always going to happen, but this is a great illustration how the once king Google Search is now falling. Sure, Google might save the situation by moving faster into AI but it will be really hard for them to kill their darlings. (Now, we just need AI that links to sources)
To illustrate the difference. Here's me asking ChatGPT, and getting the answer I needed (the GLB class, that I was trying to find).
And the first 3 pages of Google results:
Page 1: only ads
Page 2-3: zero results for Mercedes gas cars: only about electric ones. Zero!!
ALT List the 3 Mercedes gas models that are closest to the Mercdes EQB. As of my knowledge cutoff date in September 2021, the Mercedes EQB is an all-electric compact SUV that shares similarities with other compact SUVs and crossovers in the Mercedes-Benz lineup. The three closest gasoline-powered Mercedes models to the EQB are: Mercedes-Benz GLB-Class: The GLB is the most similar to the EQB, as the EQB is essentially an electric version of the GLB. The GLB is a compact SUV that offers a spacious interior and shares similar dimensions and design cues with the EQB. Mercedes-Benz GLA-Class: The GLA is a subcompact crossover SUV and is slightly smaller than the EQB. However, it still shares some similarities in terms of design, features, and target audience. Mercedes-Benz GLC-Class: The GLC is a compact luxury SUV that is slightly larger and more upscale than the EQB. It is designed to offer a balance between performance, luxury, and utility. Please note that these comparisons are based on
ALT Google search results for "Mercedes gas models that are closest to the Mercedes EQB."
ALT Google search results for "Mercedes gas models that are closest to the Mercedes EQB."
ALT Google search results for "Mercedes gas models that are closest to the Mercedes EQB."
ripperdoc retweeted
We accidentally invented computers that can lie to us and we can't figure out how to make them stop
ripperdoc retweeted
Yeah! And every seasoned programmer knows that *writing* code is trivial and what takes most of the time is figuring out the right way of doing things. That’s where the value of software engineers lie, not in typing code.
ripperdoc retweeted
đź’Ż this
Any time I want to work with a language ecosystem that's outside my direct comfort zone the hardest bit is ALWAYS bootstrapping myself to a point where I can start working with code - and feeling confident that I didn't make any weird mistakes along the way
I’ve been testing ChatGPT on a lot of my day to day coding problems. It’s uniquely suitable for coding, because almost every problem is unique in some way and therefore not trivially googled. It’s been helpful but also often hallucinates invalid code. So it’s hit and miss