Human intelligence is super-specialized for the physical world, and our feeling of generality is an illusion
We only seem general because we can't imagine the problems we're blind to
We only seem general because we can't imagine the problems we're blind to
think through first principles and figure out a novel approach which is missing here to make it train significantly faster and generalize even with this small dataset
2D Fourier Transforms https://www.cs.toronto.edu/~jepson/csc320/notes/linearFilters2.pdf
If you're 20β30:
- You probably already know what you like. The danger isn't ignorance, it's endless deferral dressed up as optionality. Pick.
- Tight feedback loops. Short cycle times. The thing that compounds is reps, not plans.
- Your taste exceeds your ability now. That's the gap. Close it with volume.
- Get on the frontier of something, even if small. See the problems firsthand. Most moats come from noticing what matters before it's obvious.
- Specific knowledge > generic credentials. The person who deeply understands how X works beats the better resume. Information is cheap now; sorting by credential is legacy behavior.
- Find your people. Collaborators, co-conspirators. Your network at 30 is largely downstream of who you meet now.
- Default to variance. 10% chance of great beats 70% chance of fine. You can absorb downside; you can't manufacture time.
- Be useful to people doing important things. Proximity to interesting problems > title.
- Your body still forgives you. It won't forever. Sleep, exercise, don't drink too much. Establish the habits you want at 50.
- Impatience with outcomes, patience with process.
Failure mode of 10β20: closing too early.
Failure mode of 20β30: staying open too long.
- You probably already know what you like. The danger isn't ignorance, it's endless deferral dressed up as optionality. Pick.
- Tight feedback loops. Short cycle times. The thing that compounds is reps, not plans.
- Your taste exceeds your ability now. That's the gap. Close it with volume.
- Get on the frontier of something, even if small. See the problems firsthand. Most moats come from noticing what matters before it's obvious.
- Specific knowledge > generic credentials. The person who deeply understands how X works beats the better resume. Information is cheap now; sorting by credential is legacy behavior.
- Find your people. Collaborators, co-conspirators. Your network at 30 is largely downstream of who you meet now.
- Default to variance. 10% chance of great beats 70% chance of fine. You can absorb downside; you can't manufacture time.
- Be useful to people doing important things. Proximity to interesting problems > title.
- Your body still forgives you. It won't forever. Sleep, exercise, don't drink too much. Establish the habits you want at 50.
- Impatience with outcomes, patience with process.
Failure mode of 10β20: closing too early.
Failure mode of 20β30: staying open too long.
North Star: systems that help people understand what's actually happening in important human interactions
π‘ Remember Box
convert taste into compounding position
good taste in where to apply effort
You woke up.
There is no death. (Bhagavad GΔ«tΔ)
There is no fear. (Dhammapada)
There is no separation. (Upanishads)
There is no agency. (Tao Te Ching)
There is no self. (PΔli Canon)
There is no permanence. (Heart SΕ«tra)
There is no wilderness but God. (Ibn ΚΏArabΔ«)
There is no veil but the veil. (Zohar)
There is no world but the Word. (Gospel of John)
There is no end, only return. (Book of Revelation)
What a day that will be.
There is no death. (Bhagavad GΔ«tΔ)
There is no fear. (Dhammapada)
There is no separation. (Upanishads)
There is no agency. (Tao Te Ching)
There is no self. (PΔli Canon)
There is no permanence. (Heart SΕ«tra)
There is no wilderness but God. (Ibn ΚΏArabΔ«)
There is no veil but the veil. (Zohar)
There is no world but the Word. (Gospel of John)
There is no end, only return. (Book of Revelation)
What a day that will be.
General Intelligence Company | Memory Is The Last Problem To Solve To Reach Agi
https://www.generalintelligencecompany.com/writing/memory-is-the-last-problem-to-solve-to-reach-agi
https://www.generalintelligencecompany.com/writing/memory-is-the-last-problem-to-solve-to-reach-agi
General Intelligence Company
Memory Is The Last Problem To Solve To Reach Agi | General Intelligence Company
Read this blog post from General Intelligence Company.
β€1
π‘ Remember Box
https://mvakde.github.io/blog/why-all-ARC-solvers-fail-today/
Why isn't image and video codecs built into models? π€ maybe some of it already do use compressed formats?
Alternatively, why isn't there a new image compression standard based on modern ml research?
Feels like low hanging fruit
Alternatively, why isn't there a new image compression standard based on modern ml research?
Feels like low hanging fruit
what have we done so far? what should we do next? how should we do it? what should we try? how can we improve how we iterate? how can we do things better?