.

TODO

@@ -0,0 +1,20 @@
+questions
+- how does algebraic statistics relate to my post about representations in linear algebra?
+- 
+
+
+
+ideas
+- what would it take to fabricate a microchip at home?
+- imagine a person who would look at my life and be very jealous
+- Diagnosing countries and their mental disorders
+- Abolish mathematics. Mathematics was invented to help rich people keep track of their assets.
+- Pair tools and mathematical theorems. The hammer: ???,
+- Open problems for my back pocket. To carry around with me all the time.
+- CAPTCHAs. history / arms race. https://github.com/jpraychev/google-recaptcha
+- Generating pseudorandom numbers. Intuition, make a chaotic system. But how do you ensure it gives a uniform distribution!?
+- how many different types of memory are there?
+- are markets really efficient? What are the alternatives?
+- the importance of good audio design?! A set of 3d scenarios. With the player turning their head / moving.
+With poor / good audio design
+- 

_drafts/inbetween-posts/2020-10-12-mechanical-components.md

@@ -3,6 +3,8 @@ layout: post
 title: Mechanical components
 ---
 
+https://www.youtube.com/watch?v=M1-YeqGynlw
+
 You may already be familiar with electrical components like; transistors, capacitors, operational-amplifiers, flip-flops, etc...  Electrical components are defined as; ...?
 
 ![]({{site.baseurl}}/assets/mechanical-components/components.jpeg)

_drafts/inbetween-posts/2024-03-10-brilliant.md

@@ -0,0 +1,22 @@
+---
+title: "Brilliant"
+subtitle: ""
+layout: post
+permalink: /brilliant/
+categories:
+    - sci-fi
+---
+
+<!-- 
+Aliens on a distant planet have a unique way of communicating. They use visual patterns generated on their bodies to convey messages. 
+The patterns can convey entire sentences in a single image. (like in the heptapods in arrival) 
+
+This planet has has frequent auroras that light up the sky.
+The aliens are deeply religious and believe that the auroras are messages from their gods.
+
+Most of the time, the auroras are random patterns. But every once in a while, the patterns form a word / sentence.
+These are the words of the gods.
+--> 
+
+The gods have not spoken in a week.
+Abas lies 

_drafts/inbetween-posts/2024-03-10-the-conductor.md

@@ -0,0 +1,19 @@
+---
+title: "The conductor"
+subtitle: "Consciousness is the conductor of the orchestra of the mind"
+layout: post
+permalink: /conductor/
+categories:
+    - philosophy
+---
+
+![Conductor]({{ site.baseurl }}/assets/conductor.jpeg)
+
+When I look at a conductor, I don't see a person in control of the orchestra. I see a delusion.
+
+It seems to me that the music would happen with or without the conductor.
+However, the conductor believes that they are in control of the music. They believe that the music would not happen without them.
+Confusing correlation with causation, they raise their hands, and the crescendo begins.
+
+I believe this is the perfect metaphor for consciousness. Like the conductor, we believe that we are in control of our thoughts and actions.
+That we are making the music that is our life.

_posts/inbetween-posts/2019-02-22-fusion-pollution.md

@@ -16,8 +16,8 @@ We achieve an astonishing fusion reaction output of 30:1 output to input, named
 
 (20 years later) A report is written on environmental impact of BARY. Li may pollute our atmosphere. Lithium is highly reactive and if released in sufficient quantities it could significantly alter the amount of oxygen in the atmosphere.
 
-(30 years later) The new norm is cheap energy. Young thrill-seekers swap their automotive rides for miniaturized spacecraft. Movies and video games are rendered in `65536p` (despite the fact we can't tell the difference between `2048p` and `4096p`). People even take 1hr long showers!
+(30 years later) The new norm is cheap energy. Young thrill-seekers swap their automotive rides for miniaturized spacecraft. Movies and video games are rendered in `65536p` (despite the fact we can't tell the difference between `2048p` and `4096p`). People even take 1hr long showers (cos who does that now???)!
 
 (50 years later) Oxygen levels are declining. And we are seeing the effects of it! Some marine biologists have noticed large populations of fish missing. Separately, entomologists are documenting a decline in ant populations! Some scientists link waste lithium from BARY to the reduction in oxygen levels, but according to others there is little consensus on the issue. The conclusion is that the climate is changing. But is it anthropogenic?
 
-Same idiots, similar story, different physics.
+Same idiots, different physics.

_posts/inbetween-posts/2020-07-03-outsourced.md

@@ -18,32 +18,32 @@ People are removing their gastrointestinal organs and buying Gastro which regula
 
 ### Physio
 
-> Exercising (or simply walking) takes attention, and attention is a valuable resource. Don't waste another glance on such unimportant tasks, let us handle it for you.
-> The act of walking, running or simply exercising require attention, a valuable resource. What if you wouldn't have to spend another moment attending to these tasks? 'Physio' has you covered.
+> The act of walking, running or exercising require attention, a valuable resource. What if you wouldn't have to spend another moment attending to these tasks? 'Physio' has you covered.
 
 When switched on: control of your limbs will be outsourced to our state-of-the-art motor control policies. They have some great features like; auto-pilot, self-preservation, efficiency, workout, ... It cant get lost, it won't injure you, it won't harm others!
 
 ### Psydy
 
-> Learning a new language, sport, ..., can be confusing, intimidaing and hardas easily as selecting the topic and waiting for a week? Well, it is that easy with Psydy!
+> Learning a new language, sport, ..., can be confusing, intimidaing and just hard.
+What if it was as easy as selecting the topic and waiting for a week? Well, it is that easy with Psydy!
 
-While you live your life as usual, we train a copy of your brain on the knowledge or skills you wish to aquire; taekwondo, violin, political science, 1920's art history, ... . We then update your brain to match the trained copy, imparting the learning done.
+While you live your life as usual, we train a copy of your brain on the knowledge or skills you wish to aquire; taekwondo, violin, political science, 1920's art history, ... . We then update your brain to include the trained copy, imparting the new skills.
 
 ### Emote
 
 > Sometimes emotions get in the way of life. Sad for no obvious reason? Hangry, but you just ate? Need to focus, now!?
 
 'Emote' offers a way to modulate our emotional states like never before.
 
-Emote is a small device that sits on your brain and modulates the activity of your amygdala. It can be used to increase or decrease the intensity of emotions. It can also be used to induce emotions, like happiness, sadness, anger, ... .
+Emote is a implant that help modulate the activity of your amygdala, hypothalamus, Pituitary gland, hippocampus, Pineal gland. It can also be used to induce emotions, like happiness, sadness, anger, ... To make you more productive, more social, more creative, ...
 
 * * *
 
 These advertisements are supposed to highlight of the dangers of outsourcing our own lives.
 
 > Every convenience a piece of our humanity relinquished?
 
-Some famous examples include;
+Inspired by the following examples of outsourcing:
 
 In the early '80s, IBM outsourced their software needs to a then-small company, Microsoft. Microsoft retain the rights to the software and sold it to other companies, effectively giving them a professional head start. This decision is hindsight is noted as one of the main factors contributing to Microsoft becoming a global tech giant. [ref](https://spectrum.ieee.org/how-the-ibm-pc-won-then-lost-the-personal-computer-market)
 

_posts/inbetween-posts/2024-08-10-utilitarianism.md

@@ -21,14 +21,16 @@ This thought experiment is used to explore the ethical dilemma of whether it is
 The correct answer appears to be to pull the lever (or transplant the organs), saving the five people at the expense of the one person, maximizing the number of lives saved.
 However, there exist many arguments against this conclusion. Each as nonsensical as the last.
 
-What is often ignored in this thought experiment is __(un)certainty__. In this case, we are certain that pulling the lever (or transplanting) will save the five people. However, in the real world, we are often uncertain about the consequences of our actions. This uncertainty is the problem with utilitarianism in practice.
+What is often ignored in this thought experiment is __(un)certainty__! In this case, we are certain that pulling the lever (or transplanting) will save the five people. However, in the real world, we are always uncertain about the consequences of our actions.
 
-We cannot predict the future. We cannot be sure that transplanting a heart will lead to a full recovery.
+Some will argue that we can work with risk. We can evaluate the probability of different outcomes and act in a way that maximizes the expected value. However, this is also not possible, for a similar reason. We can never know the true probability of different outcomes given our actions.
 
-Consider the following scenario:
+Because we cannot know the future with certainty we can never evaluate one action as better than another.
+
+<!-- Consider the following scenario:
 
 > A trolley is hurtling down a track towards five people. You are standing next to a lever that will probably divert (75%) the trolley onto a side track where one person is standing. But the train may derail and kill everyone (25%). Do you pull the lever?
 
 Again, the correct answer appears to be to pull the lever, in expectation saving 3.75 people.
 ($0.75 \times 5 + 0.25 \times 0 = 3.75 $).
-However, we have again assumed we can predict the future with certainty. Where did the 75% and 25% come from? How can we be sure that the trolley has a 75% chance of being diverted?
+However, we have again assumed we can predict the future with certainty. Where did the 75% and 25% come from? How can we be sure that the trolley has a 75% chance of being diverted? -->

_posts/personal-posts/2015-12-29-future-smarter-telf.md

@@ -1,10 +1,5 @@
 ---
 title: "Future Telf, 2016"
-date: "2015-12-29"
-tags: 
-  - "dreams"
-  - "future"
-  - "inspiration"
 layout: post
 subtitle: Dreaming about projects
 ---

_posts/inbetween-posts/2018-11-24-open-minded-a-game.md → _posts/personal-posts/2018-11-24-open-minded-a-game.md

@@ -3,7 +3,7 @@ title: "Being open minded"
 date: "2018-11-24"
 coverImage: "screenshot-2018-11-24-at-8-11-41-pm.png"
 layout: post
-subtitle: Play a 'game' with me
+subtitle: Let's take turns reading books
 categories: 
   - "interact"
 ---

_posts/personal-posts/2020-07-03-adversarial-collaboration-contest.md

@@ -1,9 +1,10 @@
 ---
 title: "Adversarial collaboration contest"
-date: "2020-07-03"
 coverImage: "woman-yelling-at-cat.jpg"
 layout: post
-subtitle: Play a game with me.
+subtitle: Let's have a tolerant debate
+categories:
+    - "interact"
 ---
 
 ![]({{site.baseurl}}/assets/adversarial-collaboration-contest/{{page.coverImage}})

_drafts/technical-posts/2022-11-10-lm-chem.md → _posts/technical-posts/2022-11-10-lm-chem.md

@@ -1,7 +1,12 @@
 ---
 layout: post
-title: Language models are all you need (for SMILES-based chemistry)
+title: Language models are all you need
+subtitle: for SMILES-based chemistry
 permalink: lm-chem
+categories:
+  - proposal
+scholar:
+  bibliography: "lm-chem.bib"
 ---
 
 > Let's train a single large language model on for as many different chemistry tasks as possible.
@@ -32,7 +37,7 @@ Or;
 - _"Molecular formula: C21H22N2O2. Observed 13C NMR peaks: 1.22, 1.41, 1.83, 1.84, ... Elucidated structure: \_\_\_\_\_\_\_\_\_\_"_<br>
 - Completion: _"O=C1CC2OCC=C3C4C2C2N1c1ccccc1C12CCN(C1C4)C3"_
 
-We could use data from [NMRShiftDB](https://nmrshiftdb.nmr.uni-koeln.de/) for this task.
+We could use data from NMRShiftDB {% cite Kuhn2015FacilitatingQC %} for this task.
 
 ### Translation from natural language to synthetic 'program'
 
@@ -45,17 +50,17 @@ A dataset for this could be sourced from [orgsyn](http://www.orgsyn.org/) or [Pi
 
 ***
 
-> All of chemistry in a LM.
-
 What are other tasks we could include?
 
 - reactivity?
 - drug design?
 - retrosynthesis?
 
-How to include other types of chemical data?
+> All of chemistry in a LLM!
+
+Future work could investigate how to include other types of chemical data. For example;
 
-- [QM9](http://quantum-machine.org/datasets/) (smiles, 3D positions, electron densities)
+- [QM9](http://quantum-machine.org/datasets/) (3D positions, electron densities)
 - [Open catalyst project](https://opencatalystproject.org/) (trajectories, forces and energies)
 - Theoretical knowledge. Like the dataset used [here](https://chemrxiv.org/engage/api-gateway/chemrxiv/assets/orp/resource/item/6393827c836cebbc757aedeb/original/assessment-of-chemistry-knowledge-in-large-language-models-that-generate-code.pdf) or what about past exams?
 
@@ -64,34 +69,59 @@ How to include other types of chemical data?
 
 Recently there have been two important observations with AI;
 
-- Deep learning <u>works</u> when done at scale. The bigger the better.
-- The ability of strings to represent many different kinds of information allows great flexibility which tasks a LM is trained to do.
+1. Deep learning works when done at scale. The bigger the better.
+2. The ability of strings to represent many different kinds of information allows great flexibility which tasks a LLM can be trained to do.
 
-This trend of BIGGER deep learning can be seen by;
+### Deep learning at scale
 
 - More parameters
-  - [Gopher](https://arxiv.org/abs/2112.11446). 280B parameters
-  - [PaLM](https://arxiv.org/abs/2204.02311). 540B parameters
-  - [Switch Transformers](https://arxiv.org/abs/2101.03961). 1T parameters
+  - Gopher {% cite Rae2021ScalingLM %}. 280B parameters
+  - PaLM {% cite Chowdhery2022PaLMSL %}. 540B parameters
+  - Switch Transformers {% cite Fedus2021SwitchTS %}. 1T parameters
 - More data
-  - [MassiveText](https://arxiv.org/abs/2112.11446). 2 trillion tokens
-  - [LTIP](https://arxiv.org/abs/2111.02114). 400 million image-caption pairs
+  - MassiveText {% cite Rae2021ScalingLM %}. 2 trillion tokens
+  - LTIP {% cite schuhmann2021laion400mopendatasetclipfiltered %}. 400 million image-caption pairs
 - More tasks
-  - A generalist agent [GATO](https://www.deepmind.com/publications/a-generalist-agent). Trained on;
+  - A generalist agent GATO {% cite reed2022generalistagent %}. Trained on;
     - Simulated control tasks (596 tasks) ([DM Control](https://github.com/deepmind/dm_control), [DM lab](https://www.deepmind.com/open-source/deepmind-lab), [Procgen](https://openai.com/blog/procgen-benchmark/), [Atari ALE](https://github.com/mgbellemare/Arcade-Learning-Environment),  [playroom](https://arxiv.org/abs/1707.03300), ... and more)
     - vision and language (>204 tasks) ([MassiveText](https://arxiv.org/abs/2112.11446), [MultiModal MassiveWeb](https://arxiv.org/abs/2204.14198), [LTIP](https://arxiv.org/abs/2111.02114), [OKVQA](https://okvqa.allenai.org/), ... and more)
 
 
-Language models
+### The flexibility of strings and power of LLMs
 
-A 'model' of language. Given context predict a distribution over the likely next token.
+A language model is a 'model' of language. It models language as a prediction problem: given context predict a distribution over the likely next token.
 
 - "The cat in the _" __->__ "hat" (probably)
 - "Monday, Tuesday, Wednesday, _" __->__ Thursday (probably)
 - "The most populous city in India is _" __->__ Mumbai (probably)
 
+Given a language model, we can frame many NLP tasks as predicting the next token, given a prompt.
+This lead some to claim _"Language models are all you need"_ (for NLP tasks) {% cite namazifar2020languagemodelneednatural %}.
 
-## Aside: Few-shot meta learning vs fine-tuning
+For example; narrative understanding, textual entailment, entity resolution, question answering, POS tagging, grammatic parsing... can all be framed as predicting the next token. And thus can be done with a LLM.
+
+- textual entailment: "text: If you help the needy, God will reward you. hypothesis: Giving money to a poor man has good consequences." -> "positive"  (text entails hypothesis)
+- POS tagging: "text: Bob made a book collector happy." -> "subject verb object(article adjective noun) verb-modifier"
+- sentiment analysis: "text: I love this movie!" -> "positive"
+- question answering: "text: The capital of France is _" -> "Paris"
+
+## Aside: more fun tasks LMs can do
+
+_Big-Bench_ {% cite srivastava2023imitationgamequantifyingextrapolating %}
+
+Analyses a LM's ability to do 204 different tasks including.
+
+- auto_debugging
+  - "'\\nfor i in range(10):\\n\\ti' What is the value of i the third time line 2 is executed?" __->__ "2"
+- color matching
+  - "What is the color most closely matching this RGB representation: rgb(128, 2, 198)?" __->__ "purple"
+- chess_state_tracking_legal_moves
+  - "e2e4 g7g6 d2d4 f8g7 c1e3 g8f6 f2f3 d7d6 d1" __->__ "c1, d2, d3, e2"
+
+others
+- ascii mnist, solve riddles, play sudoku, translate hindi proverbs, idendify math thorems, vitamin C fact verification, etc...
+
+<!-- ## Aside: Few-shot meta learning vs fine-tuning
 
 Fine-tune
 
@@ -119,31 +149,14 @@ _A: The 1992 Olympics were held in Barcelona, Spain._
 
 _Q: What is human life expectancy in the United States?_
 
-(example from openai's api)
-
-## Flexible prompt design
-
-_Language models are all you need_ (for natural language [@Namazifar2020])
-
-As far as we know, it is possible to frame any many tasks as a prompt to a LM.
-narrative understanding, textual entailment, entity resolution, question answering, POS tagging, grammatic parsing...
+(example from openai's api) -->
 
-- textual entailment: "text: If you help the needy, God will reward you. hypothesis: Giving money to a poor man has good consequences." -> "positive"  (text entails hypothesis)
-- POS tagging: "text: Bob made a book collector happy." -> "subject verb object(article adjective noun) verb-modifier"
+## Conclusion
 
-## Aside: more fun tasks LMs can do
+Molecules and analytical data can be represented as strings. This allows us to use LLMs to do chemistry tasks.
+But how can we cram enough knowledge into a LLM to make it useful for chemistry?
+Let's train a single large language model on for as many different chemistry tasks as possible.
 
-_Big-Bench_ [@Srivastava2022]
-
-Analyses a LM's ability to do 204 different tasks including.
-
-- auto_debugging
-  - "'\\nfor i in range(10):\\n\\ti' What is the value of i the third time line 2 is executed?" __->__ "2"
-- color
-  - "What is the color most closely matching this RGB representation: rgb(128, 2, 198)?" __->__ "purple"
-- chess_state_tracking_legal_moves
-  - "e2e4 g7g6 d2d4 f8g7 c1e3 g8f6 f2f3 d7d6 d1" __->__ "c1, d2, d3, e2"
-
-others
-- ascii mnist, solve riddles, play sudoku, translate hindi proverbs, idendify math thorems, vitamin C fact verification, etc...
+## Bibliography
 
+{% bibliography --cited %}

index.md

@@ -8,12 +8,11 @@ Here are some links to help you navigate:
 
 - [Personal]({{site.baseurl}}/personal/)
     - my [experiences]({{site.baseurl}}/experiences/)
-    - (my poor) [mental health]({{site.baseurl}}/mental-health/)
+    - invitations for [interaction]({{site.baseurl}}/interact/)
     - and more
 - [Opinions and thoughts]({{site.baseurl}}/inbetween/)
     - [economics and politics]({{site.baseurl}}/economics-politics/)
     - [philosophising about ...]({{site.baseurl}}/philosophy/)
-    - [with invitations for interaction]({{site.baseurl}}/interact/)
     - [speculation]({{site.baseurl}}/speculation/)
     - [sci-fi story ideas]({{site.baseurl}}/sci-fi/)
     - and more

pages/interact.md

@@ -0,0 +1,13 @@
+---
+layout: page
+title: Intract with me
+permalink: /interact/
+---
+
+<div class="posts">
+    {% for post in site.posts %}
+        {% if post.categories contains "interact"%}
+            <p><u><a href="{{ site.baseurl }}{{ post.url }}">{{ post.title }}</a></u><br><i>{{ post.subtitle }}</i></p>
+        {% endif %}
+    {% endfor %}
+<div class="posts">