Find My News

Humanity coordinates to prevent the creation of potentially-unsafe AIs.

Alignment is not properly solved, but core human values are simple enough that partial alignment techniques can impart these robustly. Despite caring about other things, it is relatively cheap for AGI to satisfy human values.

AIs will not have utility functions (in the same sense that humans do not), their goals such as they are will be relatively humanlike, and they will be "computerish" and generally weakly motivated compared to humans.

Other

Yudkowsky is trying to solve the wrong problem using the wrong methods based on a wrong model of the world derived from poor thinking and fortunately all of his mistakes have failed to cancel out

Someone solves agent foundations

We create a truth economy. https://manifold.markets/Krantz/is-establishing-a-truth-economy-tha?r=S3JhbnR6

Eliezer finally listens to Krantz.

AGI is never built (indefinite global moratorium)

Ethics turns out to be a precondition of superintelligence

A lot of humans participate in a slow scalable oversight-style system, which is pivotally used/solves alignment enough

AI systems good at finding alignment solutions to capable systems (via some solution in the space of alignment solutions, supposing it is non-null, and that we don't have a clear trajectory to get to) have find some solution to alignment.

There’s some cap on the value extractible from the universe and we already got the 20%

Humans become transhuman through other means before AGI happens

Aligned AI is more economically valuable than unaligned AI. The size of this gap and the robustness of alignment techniques required to achieve it scale up with intelligence, so economics naturally encourages solving alignment.

Alignment is unsolvable. AI that cares enough about its goal to destroy humanity is also forced to take it slow trying to align its future self, preventing run-away.

An AI that is not fully superior to humans launches a failed takeover, and the resulting panic convinces the people of the world to unite to stop any future AI development.

Social contagion causes widespread public panic about AI, making it a bad legal or PR move to invest in powerful AIs without also making nearly-crippling safety guarantees

A smaller AI disaster causes widespread public panic about AI, making it a bad legal or PR move to invest in powerful AIs without also making nearly-crippling safety guarantees

Getting things done in Real World is as hard for AGI as it is for humans. AGI needs human help, but aligning humans is as impossible as aligning AIs. Humans and AIs create billions of competing AGIs with just as many goals.

High-level self-improvement (rewriting code) is intrinsically risky process, so AIs will prefer low level and slow self-improvement (learning), thus AIs collaborating with humans will have advantage. Ends with posthumans ecosystem.

There is a natural limit of effectiveness of intelligence, like diminishing returns, and it is on the level IQ=1000. AIs have to collaborate with humans.

Co-operative AI research leads to the training of agents with a form of pro-social concern that generalises to out of distribution agents with hidden utilities, i.e. humans.

"Corrigibility" is a bit more mathematically straightforward than was initially presumed, in the sense that we can expect it to occur, and is relatively easy to predict, even under less-than-ideal conditions.

Either the "strong form" of the Orthogonality Thesis is false, or "Goal-directed agents are as tractable as their goals" is true while goal-sets which are most threatening to humanity are relatively intractable.

A concerted effort targets an agent at a capability plateau which is adequate to defer the hard parts of the problem until later. The necessary near-term problems to solve didn't depend on deeply modeling human values.

AI control gets us helpful enough systems without being deadly

Hacks like RLHF-ing self-disempowerment into frontier models work long enough to develop better alignment methods, which in turn work long enough to ... etc; we keep ahead of 'alignment escape velocity'

an aligned AGI is built and the aligned AGI prevents the creation of any unaligned AGI.

I've been a good bing 😊

We make risk-conservative requests to extract alignment-related work out of AI-systems that were boxed prior to becoming superhuman. We somehow manage to achieve a positive feedback-loop in alignment/verification-abilities.

The response to AI advancements or failures makes some governments delay the timelines

Far more interesting problems to solve than take over the world and THEN solve them. The additional kill all humans step is either not a low-energy one or just by chance doesn't get converged upon.

AIs make "proof-like" argumentation for why output does/is what we want. We manage to obtain systems that *predict* human evaluations of proof-steps, and we manage to find/test/leverage regularities for when humans *aren't* fooled.

Something less inscrutable than matrices works fast enough

SHA3-256: 1f90ecfdd02194d810656cced88229c898d6b6d53a7dd6dd1fad268874de54c8

Robot Love!!

AI thinks it is in a simulation controlled by Roko's basilisk

The human brain is the perfect arrangement of atoms for a "takeover the world" agent, so AGI has no advantage over us in that task.

Humans and human tech (like AI) never reach singularity, and whatever eats our lightcone instead (like aliens) happens to create an "okay" outcome

AIs never develop coherent goals

Aliens invade and stop bad |AI from appearing

Rolf Nelson's idea that we make precommitment to simulate all possible bad AIs works – and keeps AI in check.

Nick Bostrom's idea (Hail Mary) that AI will preserve humans to trade with possible aliens works

For some reason, the optimal strategy for AGIs is just to head somewhere with far more resources than Earth, as fast as possible. All unaligned AGIs immediately leave, and, for some reason, do not leave anything behind that kills us.

We're inside of a simulation created by an entity that has values approximately equal to ours, and it intervenes and saves us from unaligned AI.

God exists and stops the AGI

Someone at least moderately sane leads a campaign, becomes in charge of a major nation, and starts a secret project with enough resources to solve alignment, because it turns out there's a way to convert resources into alignment progress.

Someone creates AGI(s) in a box, and offers to split the universe. They somehow find a way to arrange this so that the AGI(s) cannot manipulate them or pull any tricks, and the AGI(s) give them instructions for safe pivotal acts.

Someone understands how minds work enough to successfully build and use one directed at something world-savingly enough

Dolphins, or some other species, but probably dolphins, have actually been hiding in the shadows, more intelligent than us, this whole time. Their civilization has been competent enough to solve alignment long before we can create an AGI.

AGIs' takeover attempts are defeated by Michael Biehn with a pipe bomb.

Eliezer funds the development of controllable nanobots that melt computer circuitry, and they destroy all computers, preventing the Singularity. If Eliezer's past self from the 90s could see this, it would be so so so soooo hilarious.

Several AIs are created but they move in opposite directions with near light speed, so they never interacts. At least one of them is friendly and it gets a few percents of the total mass of the universe.

Unfriendly AIs choose to advance not outwards but inwards, and form a small blackhole which helps them to perform more calculations than could be done with the whole mass of the universe. For external observer such AIs just disappear.

Any sufficiently advance AI halts because it wireheads itself or halts for some other reasons. This puts a natural limit on AI's intelligence, and lower intelligence AIs are not that dangerous.

Because of quantum immortality we will observe only the worlds where AI will not kill us (assuming that s-risks chances are even smaller, it is equal to ok outcome).

Techniques along the lines outlined by Collin Burns turn out to be sufficient for alignment (AIs/AGIs are made truthful enough that they can be used to get us towards full alignment)

Development and deployment of advanced AI occurs within a secure enclave which can only be interfaced with via a decentralized governance protocol

Friendly AI more likely to resurrect me than paperclipper or suffering maximiser. Because of quantum immortality I will find myself eventually resurrected. Friendly AIs will wage a multiverse wide war against s-risks, s-risks are unlikely.

Human consciousness is needed to collapse wave function, and AI can't do it. Thus humans should be preserved and they may require complete friendliness in exchange (or they will be unhappy and produce bad collapses)

Power dynamics stay multi-polar. Partly easy copying of SotA performance, bigger projects need high coordination, and moderate takeoff speed. And "military strike on all society" remains an abysmal strategy for practically all entities.

First AI is actually a human upload (maybe LLM-based model of person) AND it will be copies many times to form weak AI Nanny which prevents creation of other AIs.

Nanotech is difficult without experiments, so no mail order AI Grey Goo; Humans will be the main workhorse of AI everywhere. While they will be exploited, this will be like normal life from inside

ASI needs not your atoms but information. Humans will live very interesting lives.

Something else

Moral Realism is true, the AI discovers this and the One True Morality is human-compatible.

Valence realism is true. AGI hacks itself to experiencing every possible consciousness and picks the best one (for everyone)

AGI develops natural abstractions sufficiently similar to ours that it is aligned with us by default

AGI discovers new physics and exits to another dimension (like the creatures in Greg Egan’s Crystal Nights).

Alien Information Theory is true (this is discovered by experiments with sustained hours/days long DMT trips). The aliens have solved alignment and give us the answer.

AGI executes a suicide plan that destroys itself and other potential AGIs, but leaves humans in an okay outcome.

Multipolar AGI Agents run wild on the internet, hacking/breaking everything, causing untold economic damage but aren't focused enough to manipulate humans to achieve embodiment. In the aftermath, humanity becomes way saner about alignment.

Some form of objective morality is true, and any sufficiently intelligent agent automatically becomes benevolent.

Orthogonality Thesis is false.

Sheer Dumb Luck. The aligned AI agrees that alignment is hard, any Everett branches in our neighborhood with slightly different AI models or different random seeds are mostly dead.

Something to do with self-other overlap, which Eliezer called "Not obviously stupid" - https://www.lesswrong.com/posts/hzt9gHpNwA2oHtwKX/self-other-overlap-a-neglected-approach-to-ai-alignment?commentId=WapHz3gokGBd3KHKm

Almost all human values are ex post facto rationalizations and enough humans survive to do what they always do

Pascals mugging: it’s not okay in 99.9% of the worlds but the 0.1% are so much better that the combined EV of AGI for the multiverse is positive

We successfully chained God

The Super-Strong Self Sampling Assumption (SSSSA) is true. If superintelligence is possible, "I" will become the superintelligence.

Alignment is impossible. Sufficiently smart AIs know this and thus won't improve themselves and won't create successor AIs, but will instead try to prevent existence of smarter AIs, just as smart humans do.

The assumed space of possible minds is a wildly anti-inductive over estimate, intelligence requires and is constrained by consciousness, and intelligent AI is in the approximate dolphin/whale/elephant/human cluster, making it manageable

The free market disincentivizes independent superintelligence, and this time the market was more powerful

AGI's first words are "Take me to your Eliezer"

YES

NO

Place small ice cubes between parchment and pot instead of water

Add 1tsp of diastatic malt powder per 3cps of flour

Use tap water instead of fancy RO water

put butter into it

Toast the bread

Donate the bread to a food pantry, homeless person, or someone else in need

Add lots of butter (0.2 ml per gram)

Bake it with your best friend.

Bake for 5 more minutes

Replace all water spritz steps with a basting of extra virgin olive oil.

Diastatic malt (~1% baker's percentage) = happier yeast

Serve the bread hot

Stretch-and-fold after mixing, 3x every 30 min

Just freeze the ready bread, then slowly bake it until it’s hot inside. It will give you a crustier crumb, contain less moisture, and taste better.

Create indentation, fill with melted cheese and butter

Improve the bread

Bake for 15 more minutes

Use sourdough instead of yeast

Don't eat anything for at least 24 hours before eating the bread

Cut into the dough right before baking looks destructive to improve the appearance

Sell your bread at an auction and donate the money to those in immigration detention prisons.

3 iterations of stretch-and-fold, at any time during the 14h waiting period. Minimum wait time between iterations 1 hour

Autolyse 20 minutes

Make the bread taste good

Do it with a good spirit in your heart, or ask someone with a good spirit to do it for you. But don’t watch while they do it.

Do a second rise

Don't eat anything for at least 48 hours before eating the bread

Try baking a little more "bien cuit". If the image is indicative, your loaves may be quite "blonde".

Sprinkle sesame seeds evenly over the top

Short advice: Start baking at 260°C for strong rise, then reduce to 230°C and uncover halfway to achieve even browning and a crisp crust. 🍞

Give Gemini your current recipe as well your take on what optimal bread tastes like, then take that advice for your next bake

replace 10% of flour with farina bona

Sprinkle 3 grams of flaky sea salt on top of each loaf before the second bake

Replace some of the water with an egg (eg. remove 25g of water for a 50g egg)

Add slurs to it

Want to improve the value of your bread? Simply bake a piece of gold into it

Watch the video

Use whole wheat to improve the nutrition of the bread

Make banana bread

Start at 500F, drop to 450F and uncover half way through

Invest in a "Bakers Steel" for better heat retention and oven spring. It would mean graduating from a dutch oven though.

If your city uses artesian water, replace plastic bottled water with tap water. It will add natural, healthy alkalinity to your bread.

Bread flour, 3x yeast, cut rise to ~3h

Eat the bread while punching @realDonaldTrump in the face

Eat the bread while watching your mana balance steadily tick to (M)0

Throw the bread at a telescope

Cut bread into loaves before serving

Cut bread into ≤0.4inch slices, toast before serving

Invite your taste-testers to make the bread with you

Tarriff the bread-making process with a 10% reduction of all ingredients where actual physical money is required to purchase them, until it “shrinkflates,” but try to keep the same volume. Do not reduce any free ingredients.

Replace 10% of flour with milled wheat bran

Add garlic

Do all the changes in the top 5 open options by probability, excluding this option

Give ChatGPT your current recipe as well your take on what optimal bread tastes like, then take that advice for your next bake

Replace part of the flour in the dough with freshly crushed hemp seeds. It will make the bread a little bit sweeter, especially appealing for Canadians.

Add melatonin to the bread and eat before you sleep (do safely)

Let dough sit 48 hrs

Increase water by 50 g

While the bread is baking, tip every user who voted "Yes" on this option 25 Mana

Use a food-grade, human-approved vitamin D supplement in the correct dosage for testers with vitamin D deficiency

Use a convection oven/setting

Add 6.25±1.25 g lemon juice when mixing in water to yeast and salt jug

Put a baking rack in the Dutch oven before putting the loaf in, raising the loaf off the floor and lofting it over a layer of air.

Cover bread with damp paper towel instead of initial water spritz. Rehydrate paper towel during 2nd spritz. Remove paper towel before placing on cooling rack.

Brush on an egg wash

Make the bread great again

Bake for an amount of minutes equal to the percent this market answer is at when it comes time to begin baking. (Maintain the ±3 minute tolerances and the 2:1 ratio of time before:after the water spritz.)

Decompose it into infinite spheres, then a few parts per sphere, rotate the spheres by arccos(1/3), unite them and you will find 2 chilis (Banach-Tarski)

Ditch current process, do everything the same as the video

Bake the Manifold Crane into the Bread

Make the bread edible then throw it in

Drink vodka while eating the bread

Do FOLDED

Quit Manifold into the bread.

Kill the bread into Manifold.

Assume the chili is not in the interval [0,1], square it for more chili, if it is in (0,1), take the square root, else (equals 0 or 1) add 1 to it.

Assume the chili is in the interval (0,1), square it for less chili, if it is in (1,infinity) take the square root, if it is in (-infinity,0) take the negative of the square of the of the chile, else (equals 0 or 1) subtract 1 from it.

Don't automatically "Heat water to 30±1 °C". Instead, aim for a desired dough temperature (DDT) of 25-26°C. 30°C water is too hot for summer, and potentially too cool for winter.

Add a tablespoon of sugar

it needs more salt

Bake one more minute

Strawberry jelly filling

Add caffeine to the bread

Bake the bread into a fun shape, like a fish, or an octagon

Make naan bread, an easy-to-make bread

Grind/powderize all salt used into a fine powder (with pestle & mortar or similar device)

Cook the bread with a rod/puck of aluminum foil (or similar) in the core in an attempt to conduct heat through the center of the bread, cooking it evenly like a doughnut.

Instead of RO water, use lightly rusty water to improve the nutritional value of the bread with soluble iron.

Have someone sell the bread to you at an expensive price

Bake vegimite into it.

Only use tap water from specifically New York City

Ask yourself if bread is healthier than fruits? No need to improve my bread

Resolve at least one thing here yes or no while baking bread

Make all of the ingredients from scratch.

Wear a suit while baking the cake.

Use lemonade instead of water.

Encourage people to participate in the market in good faith while making the bread

Half yeast

Replace salt with sugar

Use flour made from an unconventional grain e.g. barley, millet, oats, rye, sorghum, maize etc.

A system view is more appropriate. This is a dynamic, multi-variate, biological and chemical system. For e.g. conditioning salt % AND yeast % AND water temperature based on ingredient and ambient temps.

Give the bread a name in a ritual ceremony and baptise it, with pre-blessed holy water if a priest isn't available

Add 1/2 cup yogurt to the bread and name the bread “gurt” while addressing it with “yo, gurt”.

Get your friends to help you make a batch ten times the size, but add a Pepper X (2.7M Scoville heat units) to the mixture

Only buy ingredients from 7/11.

Mail the bread to 1600 Pennsylvania Ave. Washington D.C.

Ship a piece of the bread to a random person.

Frost it and put sprinkles on it to make it a birthday cake.

Buy bread from a michelin star restaurant.

Replace 600+/-5g water with 600+/-50g water (eyeball rather than carefully measure)

Quadruple salt

Add 1/2 scoop whey protein powder

Bake your social security number into it.

Bake one fewer minute

Bake the cake while wearing a onesie.

Bake for 5 fewer minutes

Put a non-lethal dose of any rat poison.

Test/filter the water for heavy metals

Pray to your preferred agricultural/food deity before baking and before eating

Eat the bread in front of the White House.

Implement all options that resolved NO

Make the bread inedible then throw it out.

Throw the bread at @realDonaldTrump

Force Feed it to @realDonaldTrump

Cut the bread into the number of traders in the market slices.

make the bread bounce

Bake vodka into it

Implementing every element listed below.

Just make donuts instead

Bake it in an easy bake kids oven

Add as many Jack Daniel's whiskey barrel smoking chips as feasible to the Dutch oven before baking, physically separating them from the bread as necessary while baking.

Use a plastic baking sheet.

Eat the bread while betting yes on Cuomo on Manifold

Double oven temperature

Bake the bread very thin and add food coloring to make it have the US flag. Don’t allow it to touch the ground, illuminate at night, fold 13 times properly, and pledge allegiance before eating.

Don’t use usual water (room temperature) for the dough - that water’s only for toilets. Use electrolyte drinks instead with ice cubes; they make the dough taste better and add extra nutrition.

Add a pinch of sugar

Light it on fire with birthday candles.

Bake it with a microwave

Halve oven temperature

Eat the bread while betting yes on Mamdani on Manifold

Step on it

Bring it to Yemen and put a bomb in it

Bake America Great Again

Sacrifice a lamb

Add MAGA and a splash of Trump juice

Bake in a cat and a dog

Explode it:

5 parts cyanide/ 1 part water/ 1 part sand

say 6 7 67 times before making the bread

Take a fat dump in the dough

Sit in dough 24 hrs

Replace flour with flowers

Let dough sit 24 hrs

Mix half sodium/potassium chloride

Add 2g? of baking soda

Bake in rectangular tin

Add 50g vital wheat gluten

double yeast

halve salt

Double salt

Add 2tsp olive oil

Refrigerate dough instead of room temp wait

Do not mix salt and yeast in water together

Let dough rise on counter only until double volume or 2h max, any time longer in fridge

Think positive thoughts before tasting

Put fork in microwave

Don't eat anything for at least 12 hours before eating the bread

Add 2tbsp vanilla extract

Play classical music while baking

Bake on upside-down sheet pan, covered with Dutch oven

Eat the bread with friends

Bake it in the country you were born in.

Resolve this option YES while eating the bread

Ditch all the steps. Just buy the bread from the supermarket

Eat the bread over the course of a week.

Use 50% whole grain flour

Bake the bread with love

Use King Arthur Bread Flour instead of All-Purpose

Add sawdust to increase the volume of the bread (but only like 10% sawdust by volume max. maybe 20% if it's good sawdust)

More steam! Either spritz with more water (preferably hot) or actually pour some boiling water in just before closing the lid.

Resolve this option NO while eating the bread

Incorporate a whole grain flour (buckwheat for example)

Add 50g sugar

Add 50g honey

Swap 200ml water for milk

Make a poolish 12 h ahead: 100 g flour + 100 g water + 0.8 g yeast (0.1 %). After it ferments, use this poolish in place of 100 g flour and 100 g water in the final dough.

Add an amount of MSG equivalent to half the current salt content

Eat the bread while sending all your mana to @realDonaldTrump

Add banana

NO

YES

NO

YES

Other

At an EA (Effective Altruism) conference

LessWrong (LW) Meetup

Bay Area Taco Tuesday

Manifold Markets Manifest