[HN Gopher] OpenAI - transformer debugger release
___________________________________________________________________
OpenAI - transformer debugger release
Author : nmca
Score : 333 points
Date : 2024-03-12 01:12 UTC (21 hours ago)
(HTM) web link (github.com)
(TXT) w3m dump (github.com)
| CityOfThrowaway wrote:
| This is pretty cool. I find it intriguing that we're neural
| surgery on LLMs!
| seanieb wrote:
| Can't help but think Elons lawsuit will trigger more releases by
| OpenAI. His core arguments are BS, but raised legitimate
| questions about their non-profit status and lack of non-profit
| related activity.
| gurgunday wrote:
| Why do you think they are BS? You say it raised legitimate
| questions about their non-profit status and lack of non-profit
| related activity.
| rvz wrote:
| > Why do you think they are BS?
|
| _Because me and everyone else here really hates Elon Musk._
|
| > You say it raised legitimate questions about their non-
| profit status and lack of non-profit related activity.
|
| As soon as OpenAI accepted that investment from Microsoft it
| has only been used as a for-profit proxy to have their own
| 'DeepMind' to bring down Google which they have tried for
| years with Bing and the Scroogled campaigns.
|
| OpenAI is now no better than Google DeepMind and both are as
| closed and secretive and OpenAI ditched from their charter
| and non-profit status and continues to make excuses for not
| giving back their research.
| marknutter wrote:
| > Because me and everyone else here really hates Elon Musk.
|
| Speak for yourself.
| myko wrote:
| I think some believe Elon's argument is BS because Elon
| claims to be upset about OpenAI charging for their work.
| However, internal communications/texts show he wanted to do
| the same thing. So it seems like sour grapes that he isn't
| the one in charge.
| ramblerman wrote:
| An argument isn't good or bad because of the motivations of
| the person behind it.
|
| I get Elon is not popular right now. But jeez this is
| playground emotional level logic.
| _diyar wrote:
| I detailed my understanding of this case in a sibling-
| comment to this one, but I think the fact that he wanted to
| "do the same thing" (read: use the for-profit engine to
| fund AI research) only makes his case stronger!
|
| If I understand correctly, the OpenAI board asked Musk to
| remove himself and sell his stake as his for-profit
| pursuits conflicted with the mission of open, non-profit
| research. But after he left, they started a for-profit
| subsidiary and did exactly what they didn't want him doing.
| I could see how a judge might side with him.
| GreedClarifies wrote:
| Let's be real. They think Elon's arguments are BS since he
| bought Twitter and he destroyed their Leftie safe space.
|
| They are also very upset with Elon's very public shift away
| from the current Left.
|
| It's Elon derangement syndrome.
| nonethewiser wrote:
| > His core arguments are BS, but raised legitimate questions
| about their non-profit status and lack of non-profit related
| activity.
|
| These legitimate questions about their non-profit status ARE
| his core arguments. His complaints are the same as the popular
| notion on hackernews that OpenAI has failed to be open.
| refulgentis wrote:
| Erm, you get it, it's right in your comment: lawsuit about
| non-profit, HN about open weights.
| jhonof wrote:
| They are part of his core argument, but not the entire
| argument. The (imo) shakier part of the argument is that he
| is entitled to damages even though he doesn't own shares in
| the company. This is atypical, most of the time you lose that
| right once you get rid of your shares. It seems like he is
| arguing that he was misled and sold his shares because of
| that, which will be hard to prove if true, and even if it is
| true, isn't really illegal.
| _diyar wrote:
| I believe at the time it was reported that Musk and the
| OpenAI board chose to part ways due to his for-profit AI
| research (in Tesla) posing a conflict of interest to the
| OpenAI mission.
|
| Hence his argument boils down to "You made me sell my
| shares because my 'closed' AI research conflicted with your
| 'open' non-profit ideals. However, you have since stoped to
| adhere to your mission statement, effectively seizing to
| publish your research findings and pursuing for-profit
| goals. As a consequence of this I have lost a bunch of
| money."
|
| And as nuts as Musk is, I kind of see his point here.
| jhonof wrote:
| That argument doesn't really hold a lot of water with me
| to be honest. If I sell (forced or otherwise) my shares
| in company because it isn't working on tech I agree with,
| and then they pivot and start working on tech I agree
| with and their shares pop, I am not entitled to sue them
| for damages.
| _diyar wrote:
| In the scenario you described above, you would be
| entitled to sue for damages if you sold shares in a
| company under false pretenses because the other owners
| deceived you.
|
| Imagine being the part-owner of a bread focused bakery.
| You tell the other owners that you think the business
| should focus on pastries - even if everybody agrees that
| bread products are more important - as pastries make more
| money (that money can in turn be invested into the bread
| business).
|
| The other owners hard disagree and ask you to leave the
| company, because a) that does not fit into their bread-
| centric mission and b) you own a pastry-supplier in town
| and would obviously be profiting from this move. So you
| sell your shares and move on, no harm no foul.
|
| But what if they then turn around and start producing
| pastries, claiming it's only doing so to financially
| support their bread business? What if they start entering
| lucrative deals with other pastry-suppliers and
| effectively stop making bread?
|
| In this case I would argue that you would be entitled to
| some portion of your foregone gains. You sold your stock
| while being under the impression that the company would
| never go into the direction which you deem is the *only
| viable one*, only to find out that they did exactly what
| you suggested after you left.
|
| > ... because it isn't working on tech I agree with, ...
|
| Please notice that this is not what happened. They all
| agreed on the tech to be researched. The only
| disagreement was the business model and company structure
| required to fund the research.
| John23832 wrote:
| > These legitimate questions about their non-profit status
| ARE his core arguments.
|
| Except for the portion where he advocated for a capped-profit
| arm which he would run under Tesla. You know that part that
| is shown in the OpenAI emails. It seemed he was ok with it
| when HE would stand to benefit.
|
| You can't believe that Elon is "a great businessman" while
| also believing that he is totally altruistic and ignores all
| of the self serving impulses that have gotten him to where he
| is.
| revscat wrote:
| GP is discussing legal arguments, not motivations.
| John23832 wrote:
| While I am not a lawyer, I think it's pretty apparent you
| can't sue someone for damages for participating in an
| action that you conspired to be a part of.
| resource_waste wrote:
| When I said this last year, people said
|
| "BUT THEY ARE DEMOCRAITIZING IT"
|
| Its good to see that I was always right, even last year. I just
| didn't have a dad who owned an emerald mine.
| simondotau wrote:
| People seem to imagine that Errol Musk is dead and Elon got a
| big inheritance from his daddy's estate. In fact Errol is
| still very much alive, became bankrupt sometime in the 1990s,
| and is now a deadbeat living off financial support from his
| sons.
|
| Was there ever an emerald mine? Maybe. Though nobody has been
| able to find a scrap of evidence for it. Literally the only
| "evidence" is Errol's own boasts, which is that he bought a
| half-share of a Zambian mine for 100k ZAR ($50k USD). Hardly
| the stuff generational empires are made of.
| garyfirestorm wrote:
| if you bought 50k USD worth of $AAPL in 1990, by
| conservative estimates they will be worth 560 million USD
| worth today. I'd argue, that's generational stuff right
| there.
| bitcurious wrote:
| If you bought the winning lottery ticket, you'd be a
| lottery winner.
| zingelshuher wrote:
| If grandmother had balls she would be a grandfather.
| simondotau wrote:
| Did Errol Musk buy 50k worth of $AAPL in 1990?
| n2d4 wrote:
| The Zambian mine _is_ the emerald mine everyone is talking
| about. From Walter Isaacson 's biography:
|
| _> [Errol and a Zambian busienessman] agreed on a price
| [for Errol 's plane], and instead of taking a payment in
| cash, Errol was given a portion of the emeralds produced at
| three small mines that the entrepreneur owned in Zambia.
| [...]_
|
| _> Errol, who never had an ownership stake in the mine,
| expanded his trade by importing raw emeralds and having
| them cut in Johannesburg. "Many people came to me with
| stolen parcels," he says. "On trips overseas I would sell
| emeralds to jewelers. It was a cloak-and-dagger thing,
| because none of it was legal." After producing profits of
| roughly $210,000, his emerald business collapsed in the
| 1980s when the Russians created an artificial emerald in
| the lab. He lost all of his emerald earnings._
|
| $210,000 in 1986 would be about $600,000 today.
| simondotau wrote:
| Yes, all entirely consistent with what I said. Errol has
| told this story many times, including to Walter. But
| that's all there is -- a story. Is it true? Doesn't seem
| particularly consequential even if true. $600k is a lot
| of money, but it's hardly unimaginable wealth.
| brucethemoose2 wrote:
| My concern is that Elon has shut the door with a sloppy
| lawsuit.
|
| If some other entity (Government agency? Another company?)
| Comes around to pull this string, its going to set a bad
| precedent.
| furyofantares wrote:
| > raised legitimate questions about their non-profit status and
| lack of non-profit related activity
|
| What is "non-profit related activity"? IANAL but the only
| justification needed for non-profit designation is not being
| for-profit.
|
| And maybe their structure with a for-profit subsidiary is up
| for debate because of this. But you don't get to be non-profit
| by doing "good stuff" like open sourcing things. You get it by
| investing all profits back into the business.
|
| Evil Corp for Doing Evil could be a non-profit if it invested
| all its profits back into the business.
| LordDragonfang wrote:
| > What is "non-profit related activity"? What is "non-profit
| related activity"? IANAL but the only justification needed
| for non-profit designation is not being for-profit.
|
| Good thing you are NAL, because this is flat out wrong. "Non-
| profit" is a term used to refer to specific tax-exempt
| institutions under US tax law 501(c), and has nothing to do
| with investing profits "back into the business" (otherwise,
| you could have called Amazon or Uber "non-profits").
|
| OpenAI[1] is specifically a 501(c)(3)[2] which explicitly
| requires the organization's purpose be one of the following
| "exempt" purposes:
|
| >charitable, religious, educational, scientific, literary,
| testing for public safety, fostering national or
| international amateur sports competition, and preventing
| cruelty to children or animals. [3]
|
| If it does not continue to fill one of those purposes, it can
| lose its tax-exempt nonprofit status.
|
| [1] https://projects.propublica.org/nonprofits/organizations/
| 810...
|
| [2] https://www.irs.gov/charities-non-profits/charitable-
| organiz...
|
| [3] https://www.irs.gov/charities-non-profits/charitable-
| organiz...
| furyofantares wrote:
| > Good thing you are NAL
|
| I'm offended. Presumably if I was AL I would have gotten it
| right.
| jeanlucas wrote:
| Remember, open source is just on the beginning and for
| recruiting purposes
| xcodevn wrote:
| I must say, understanding how transformers work is arguably the
| most important research problem in history, assuming that AGI can
| be achieved by just scaling up current LLM models on text, video,
| audio, etc.
| kindking wrote:
| That is a very big assumption.
| ImHereToVote wrote:
| Which might turn out to be correct. Might be wrong also. We
| have no priors to AGI developing. Only NGI, and we know
| preciously little about how to achieve NGI too, except the
| bedroom way.
| joaogui1 wrote:
| In vitro fertilization too!
| furyofantares wrote:
| We have a lot of priors - everything we've ever done has
| not produced AGI.
|
| Maybe scaling transformers is the best way forward. I'm
| hopeful. But it's a big assumption that it will produce
| AGI.
| kromem wrote:
| It really depends on the definition.
|
| "Better than the _average_ human at _most_ profitable tasks "
| is a much lower bar than most people on HN might think.
|
| I have vendors who instead of filling out a web form which
| remembers their inputs and eventually even fills everything
| out for them instead print it out and fax it back in.
|
| We're probably only about 2-3 years away from transformers
| being self-optimizing enough in prompts and evaluations to
| outpace the _average_ worker in most tasks in most roles. (It
| won 't necessarily be that much cheaper after the multiple
| passes and context windows required, and crucially probably
| won't be better at _all_ tasks in most roles.)
|
| If you define AGI as "better than _any_ human at profitable
| tasks " or "better than average at _all_ tasks " then yes,
| we're a long ways off and transformers alone probably won't
| get us there.
| shafyy wrote:
| > _" Better than the average human at most profitable
| tasks"_
|
| This is not the definition of AGI. You can't just make up a
| random definition to fit your argument, lol.
| exe34 wrote:
| No that's the economically dominating definition. The
| philosophical one will happen much later or may never
| happen, but human society may change beyond recognition
| with the first one alone.
| hnben wrote:
| "The philosophical one" seems to get updated with every
| new breakthrough. 20 years ago, GPT3 would have been
| considered AGI (or "strong AI", as we called it back
| then).
|
| https://en.wikipedia.org/wiki/Artificial_general_intellig
| enc...
| exe34 wrote:
| Dennett describes it as real magic. The magic that can be
| performed is not considered real magic (it's merely a
| trick of confidence), whereas real magic is that which
| couldn't possibly be done.
| falcor84 wrote:
| I'm actually all in on people making up new definitions
| for vague terms at the start of an argument as long as
| they're explicit about it.
|
| And I particularly like this one, which is much more
| clearly measurable. If you feel AGI is taken, maybe we
| should coin this one as APGI or something like that
| worldsayshi wrote:
| I don't think the main intention was to define AGI but to
| zoom in on an interpretation of AGI that would provide
| enough value to be revolutionary.
| johnthewise wrote:
| There is no single definition of AGI. Performing most
| intellectual tasks human perform today is both general
| and a form of intelligence, so I too agree with it.
| ibejoeb wrote:
| What is _the_ definition? The definition of AGI is one of
| the central points of contention in the biggest industry
| legal battle.
| SilverBirch wrote:
| It's only in contention because 1 of the sides has a
| tonne of money, a hurt ego, and is willing to pay lawyers
| to argue the sky is red in order to get revenge on his
| former colleagues. I don't think anyone would seriously
| claim OpenAI has achieved AGI today.
| exe34 wrote:
| No, what they have is several narrow ASIs.
| belter wrote:
| This is the correct way to approach the answer to when
| and how to achieve AGI. Otherwise please present here
| your Engineer Specification for defining AGI...
|
| On a timeframe for achieving AGI:
| https://youtu.be/cEg8cOx7UZk?t=1658
| infecto wrote:
| My understanding is that AGI has no formal definition as
| it means different things to different people.
|
| The poster here created his own definition, but what is
| wrong with that? He set a very specific bar to achieve,
| something that most "high-level" thinkers in the space
| have not really done. Isn't the point of discourse to
| bring your ideas to the table?
| lm28469 wrote:
| > "Better than the average human at most profitable tasks"
|
| I think the HN crowd forgets that what really runs the
| world are min wage workers running around and doing real
| world things, not code monkeys and glorified type writers
| filling excel sheets. So yes, replacing the bullshit jobs
| we invented to keep people busy will be relatively easy,
| that's if you don't account for the fact you'll now have to
| create bullshit+ jobs to keep them busy
|
| And even then we're far away, sure it can shit out code for
| a todo webapp and create semi realistic images of a monkey
| eating a burrito but that's about it. More than a year ago
| someone betted against me here that chatgpt would
| revolutionise the world in the next year, nothing happened
| really, geeks are excited, execs are buying the hype, tons
| of money is transferred, yet there was no 4th industrial
| revolution.
|
| What happened though is that the web is flooded with
| absolutely useless content, amazon is full of ai generated
| books, students rely more and more on chatgpt to generate
| homeworks, thesis, "find" solutions, &c. it might very well
| end up being a net negative for the average joe in the long
| run
| hotdogscout wrote:
| >I think the HN crowd forgets that what really runs the
| world are min wage workers running around and doing real
| world things, not code monkeys and glorified type writers
| filling excel sheets.
|
| This is not true at all. How many products do you use
| that come primarily from minimum wage workers?
|
| If a few people responsible for Google maps running
| stopped working the GDP loss would be much bigger than if
| magnitudes more minimum wage workers did the same.
| diggan wrote:
| Forget "how many", without those people, you'd end up
| without food on your table. Then the rest of the products
| you use wouldn't matter.
| hotdogscout wrote:
| Also not true, I know the big farms of my state and they
| don't depend on minimum wage workers and I live in a
| third world country.
|
| I doubt agricultural wages or truck drivers are minimum
| wage jobs where you live. Assuming US https://www.nass.us
| da.gov/Charts_and_Maps/Farm_Labor/fl_allw...
| lm28469 wrote:
| France is one of if not the biggest agriculture power in
| Europe, most farmers can't even generate a 35hr min wage
| equivalent while working 80+ hours a week.
|
| 20% of them live in poverty, half of them make less than
| 22k euros a year
|
| Truck drivers earn between min wage and 150% of min wage,
| while being on the road every day and not having a social
| life, they drive 8 hours per day and sleep in their
| fucking truck while some code monkey makes 300k+/year
| coding memeojis at apple. Guess which ones will be
| automated first by openai lmao
| qvrjuec wrote:
| >Truck drivers earn between min wage and 150% of min wage
|
| Where are you getting this information? It's absolutely
| wrong. Long haul truckers (the one's you're saying don't
| have social lives because they drive 8 hours per day)
| make $71,196 on average in the US[1].
|
| [1] https://www.ziprecruiter.com/Salaries/LONG-HAUL-
| Truck-Driver...
| allendoerfer wrote:
| He is talking about France in the sentence before. There
| are barely any truckers in Germany with a German
| nationality. They are simply not competitive. Same goes
| for package delivery.
|
| Just imagine what would happen to a trucker's salary in
| the US if it were to create a unified market with Mexico
| and all of Central America.
|
| https://en.wikipedia.org/wiki/2004_enlargement_of_the_Eur
| ope...
|
| It's not necessarily a bad thing. Economies of Eastern
| European countries have been growing after all and
| Western Europe does not have enough workers because of
| its demographics anyway. My take is, that everybody is
| winning, there is less poverty than before, but some
| sideffects look ugly for a while.
| hotdogscout wrote:
| >half of them make less than 22k euros a year
|
| I'd be extremely happy making this amount. Some people
| are just accustomed to easier lives.
|
| >while some code monkey makes 300k+/year coding memeojis
| at apple.
|
| Meme position for a privileged caste statistically
| irrespective of skill in an institution that can piss
| money on anything and still succeed.
| diggan wrote:
| Farm work, especially work that doesn't require
| specialization (planting, maintaining, harvesting), is
| pretty much minimum wage work where I live, in Spain.
| Minimum wage here is ~1300 EUR / month. But it also
| differs wildly by region here, as some regions are really
| poor while others rich (relatively).
|
| Besides the farm work, there is food processing workers
| (cutting, cleaning, basically assembly lines), packaging,
| workers at warehouses, people who work at the counters of
| the store, and all the support roles for those positions.
| If you go outside and eat, you have all the restaurant
| personnel to take into account as well.
|
| There is a lot of low skilled labor that goes into making
| what we eat today. I'm not sure how you could possibly
| claim that none of those people are on minimum wage.
| hotdogscout wrote:
| Not all of the work you cited is essential. Would society
| crumble without retail?
|
| Minimum wage in Spain is significantly more money than
| anything I've made in my life. It's a very comfortable
| position for the vast majority of the world.
|
| >There is a lot of low skilled labor that goes into
| making what we eat today. I'm not sure how you could
| possibly claim that none of those people are on minimum
| wage.
|
| People doing essential work that isn't trivially
| replaceable have the bargaining power to charge more than
| the minimum wage in a moderately free market of human
| work, usually they do.
| squigz wrote:
| > Not all of the work you cited is essential. Would
| society crumble without retail?
|
| Did I miss the part where the other comment mentioned
| retail, or where you respond to the half dozen other
| examples of essential work?
|
| > Minimum wage in Spain is significantly more money than
| anything I've made in my life. It's a very comfortable
| position for the vast majority of the world.
|
| Instead of moving the bar some more, could you just
| define what minimum wage would be an acceptable bar for
| you in this conversation?
| hotdogscout wrote:
| Yes you missed retail, read it again.
|
| https://uk.indeed.com/career/warehouse-worker/salaries
|
| Do you really need me to Google every single essential
| position known before conceding that society is not
| maintained by minimum wage workers?
| squigz wrote:
| No, but I sure would like it if you defined it a bit
| better, since apparently we're now talking about UK
| numbers.
| CamperBob2 wrote:
| Take a look at what happened to farm employment figures
| over the last 100 years.
|
| It was a _good_ thing.
| lm28469 wrote:
| The people picking trash in my street stopped working for
| 2 days and it looks like I live in some third world
| country now, two fucking days and it looks like I live in
| the middle of an open air dump
|
| If trucks stopped deliveries every city would die in a
| week
|
| If construction workers stopped building / maintaining
| we'd be in deep shit in 6 months or less
|
| If the people in warehouses stopped working for a week
| the economy would tank like it rarely does
|
| Nurses, doctors, bus/tram/train drivers, police,
| firefighters, ambulances, janitors, trash pickers,
| plumbers, sewer workers, electricians, people taking care
| of water treatment plants, power plants, teachers, social
| workers, ...
|
| You could delete facebook, openai, instagram, twitter,
| netflix, tesla and 90% of startups from the face of the
| earth right now and I'd have the exact same life as
| yesterday. Remove any of the people I mentioned above and
| society would crumble in no time
|
| And none of these are even remotely close to being
| automated at all, nobody cares about most of these jobs.
| But hey, here is a dancing kangaroo:
| https://www.youtube.com/watch?v=Zuivg5rz_aA
| hotdogscout wrote:
| Are any of the positions you cited minimum wage workers
| where you live? Again, assuming US:
|
| https://money.cnn.com/2016/02/24/news/economy/trash-
| workers-...
|
| https://money.usnews.com/careers/best-jobs/garbage-
| collector...
|
| >You could delete facebook, openai, instagram, twitter,
| netflix, tesla and 90% of startups from the face of the
| earth right now and I'd have the exact same life as
| yesterday. Remove any of the people I mentioned above and
| society would crumble in no time
|
| Yes because you picked non-essential work. (?)
| rvnx wrote:
| Even further, maybe the world would actually be better
| without these companies.
|
| Now that there are great inventions like TikTok,
| teenagers are depressed as hell, and they don't go to
| meet each other to play soccer together, because the
| "social" networks are giving the illusion of having that
| connection.
| lm28469 wrote:
| > Yes because you picked non-essential work. (?)
|
| Then again tell me who we're automating out of the work
| force right now ? Trash pickers or code monkeys ? Truck
| drivers or artists ?
| hotdogscout wrote:
| Essential and non-replaceable are different concepts.
| fkyoureadthedoc wrote:
| All of them? We've been working on and succeeding at
| automating physical tasks for decades.
| evilduck wrote:
| Growing up my trash was picked up by a human and the
| truck crew had two or three people on it jogging house to
| house to pick up trash as the driver slow rolled through
| the neighborhood.
|
| Now my trash is serviced by one person who mostly never
| leaves the cab and who would be better described as a
| skilled machine operator than as a menial labor role. The
| work isn't completely automated but technology has
| reduced the job market for trash truck crews by two-
| thirds. I'm guessing the barrier is higher now too,
| requiring training and certifications to run the robotics
| on the truck instead of physical fitness being the
| primary prior qualification.
| _Algernon_ wrote:
| >there was no 4th industrial revolution.
|
| _Yet_. The industrial revolution didn 't happen in a
| year either.
| ecoquant wrote:
| This all reminds me of my asshole great uncle making fun
| of me as a teenager circa 1997 while I was at the
| computer and on the internet.
|
| Sarcastically asking me "Can you meet girls on that?" and
| then laughed.
|
| He wasn't wrong in the short term but laughably wrong in
| the long term.
| joenot443 wrote:
| > I think the HN crowd forgets that what really runs the
| world are min wage workers running around and doing real
| world things
|
| Is this really true? It's certainly a nice soundbyte when
| you're making class arguments or trying to dunk on the
| "HN crowd", but I think it falls apart under any level of
| scrutiny.
|
| Who keeps your lights on? Who drives your bus? Who
| manages your sewage? Who teaches your kids? Who builds
| your roads? None of them make minimum wage and would
| probably be a little insulted to be characterized as
| such.
|
| It's pretty reductionist to call anyone outside our realm
| of tech a "min wage worker", they're just workers like or
| I. I think it's a pretty stupid and pointless exercise to
| subdivide people into useful or non-useful workers,
| serving no purpose but to further pet the smugness of HN
| AI skeptics.
| weakfish wrote:
| I think this comment focuses too much on the "minimum
| wage" aspect - the core of the argument is that those are
| roles not at risk to AI in its present state, not
| necessarily the compensation aspect
| CamperBob2 wrote:
| "More than a year ago"? Really? What did anyone think was
| going to happen in a year?
|
| This sort of thing usually takes longer than you expect
| it to, and then it usually happens faster than you expect
| it to.
| Buttons840 wrote:
| > I have vendors who instead of filling out a web form
| which remembers their inputs and eventually even fills
| everything out for them instead print it out and fax it
| back in.
|
| Somewhere along the way we built computer that are so
| intuitive people find printing and faxing easier than our
| web apps. This isn't completely the fault of any single web
| app, users have a lot of learned avoidance because of bad
| experiences with many apps.
|
| In the end completely automating the job ended up being
| easier than building a good interface for a human to do the
| job.
| goatlover wrote:
| Robotics is more important to AGI, because the bulk of
| human intelligence comes from manipulating and navigating
| the physical world. Which includes a large amount of social
| interaction. Allan's are tools to assist humans. They
| aren't automating most jobs away anytime soon.
| Jensson wrote:
| Need to be better than an average expert. Humans are
| general intelligences since you can train a human to do
| anything, so a general intelligent machine needs to be able
| to be trained and become equal to human experts, matching
| an average untrained human isn't worth much.
| logicchains wrote:
| We already understand how transformers work: their architecture
| can learn to approximate a very large class of functions-on-
| sequences (specifically continuous sequence-to-sequence
| functions with compact support:
| https://arxiv.org/abs/1912.10077). It can do it more accurately
| than previous architectures like RNNs because transformers
| don't "forget" any information from prior items in the
| sequence. Training transformers to predict the next item in
| sequences eventually forces them to learn a function that
| approximates a world model (or at least a model of how the
| world behaves in the training text/data), and if they're large
| enough and trained with enough data then this world model is
| accurate enough for them to be useful for us.
|
| If you're asking for understanding the actual internal world
| model they develop, it's basically equivalent to trying to
| understand a human brain's internal world model by analysing
| its neurons and how they fire.
| theGnuMe wrote:
| >Training transformers to predict the next item in sequences
| eventually forces them to learn a function that approximates
| a world model
|
| There is absolutely no proof for this statement.
| H8crilA wrote:
| Have you used ChatGPT? It can do (at least) simple
| reasoning, for example simple spatial reasoning or simple
| human behavior reasoning.
| theGnuMe wrote:
| I suggest you start here:
|
| https://ahtiahde.medium.com/limits-of-turing-machines-
| and-al...
|
| If you don't have a CS background, I would suggest
| reading the wikipedia entries referenced in the medium
| article as well.
| seydor wrote:
| We shouldn't assume that either of those tasks is impossible
| zamfi wrote:
| > it's basically equivalent to trying to understand a human
| brain's internal world model by analysing its neurons and how
| they fire
|
| A major challenge here is that it's very hard/expensive to
| analyze these neurons, especially at any kind of scale within
| one human.
|
| Not so with LLMs.
| reexpressionist wrote:
| We may well look back in future years and view the underlying
| approach introduced in Reexpress as among the more significant
| results of the first quarter of the 21st century. With
| Reexpress, we can generate reliable probability estimates over
| high-dimensional objects (e.g., LLMs), including in the
| presence of a non-trivial subset of distribution shifts seen in
| practice. A non-vacuous argument can be made that this solves
| the alignment/super-alignment problem (the ultimate goal of the
| line of work in the post above, and why I mention this here),
| because we can achieve this behavior via composition with
| networks of arbitrary size.
|
| Because the parameters of the large neural networks are non-
| identifiable (in the statistical sense), we operate at the unit
| of analysis of labeled examples/exemplars (i.e., the observable
| data), with a direct connection between the Training set and
| the Calibration set.
|
| This has important practical implications. It works with
| essentially any generative AI model. For example, we can build
| an 'uncertainty-aware GPT-4' for use in enterprise and
| professional settings, such as law:
| https://github.com/ReexpressAI/Example_Data/blob/main/tutori...
|
| (The need for reliable, controllable estimates is critical
| regardless of any notion of AGI, since the existing LLMs are
| already getting baked into higher-risk settings, such as
| medicine, finance, and law.)
| infecto wrote:
| I hope you are not part of the founding team but if you are,
| you truly are doing your startup a disservice. Sharing your
| startup/ideas is great but doing it in the form of an
| advertisement "underlying approach introduced in Reexpress as
| among the more significant results of the first quarter of
| the 21st century" is just weird.
| pfd1986 wrote:
| Given his username I'd say.. he is.
|
| Agreed: disclaimer needed.
| blackoil wrote:
| Yeah, I am all in for hustling but this post is way over
| the top, particularly for this forum.
| 12345hn6789 wrote:
| Re: this is an advertisement for a product by its own
| employees.
| xxs wrote:
| >most important research problem in history
|
| That has to be some extremely narrowed version of all research
| that has happened (or will happen?)
| seydor wrote:
| Considering that knowing how knowing works is at the top of
| the ordo cognoscendi, it s not that narrow.
| resource_waste wrote:
| >assuming that AGI can be achieved by just scaling up current
| LLM models on text, video, audio, etc.
|
| Is any sane person actually trying this?
|
| I can't imagine an LLM ever going AGI.
| otabdeveloper4 wrote:
| > the most important research problem in history
|
| Probably not.
|
| > assuming that AGI can be achieved by just scaling up current
| LLM models
|
| Lmao.
| xcodevn wrote:
| > Probably not.
|
| Lmao.
| paulcjh wrote:
| A very meagre attempt to look like they provide open source tools
| help the world safely make AGI
| bilsbie wrote:
| How many transformers are in a typical LLM? Or is the whole thing
| considered a transformer?
| dfgtyu65r wrote:
| Normally, the LLM is composed of multiple transformer blocks,
| where each block consists of the (mutli-head) attention and
| fully-connected feedforward components. These are then stacked
| on top of each other to give the final output of the network.
| jerpint wrote:
| The architecture as a whole is referred to as a transformer
| (autoregressive decoder-only architecture in the case of gpt
| style LLMs).
|
| Note that there can be other types of LLMs too that are not
| necessarily transformer based (SSM, RNN, etc)
| nuz wrote:
| Yearly obligatory open source drop. What was it last time,
| whisper?
| whalesalad wrote:
| Interesting to see the use of ruff and black in the same project.
| https://github.com/openai/transformer-debugger/blob/main/.pr...
| jolux wrote:
| And isort, which Ruff also supports.
| throw310822 wrote:
| Hmmm.. but what happens if you allow the LLM to access and query
| its own debugger? "Why did I answer this to that?" "What happens
| if I change slightly my assumptions"?
___________________________________________________________________
(page generated 2024-03-12 23:02 UTC)