pffft8888

I was just talking about this the other day:

> it's more hacking than crareful and well specified engineering, and that could lead down a path of instability in the product where some features get better while others get worse, without understanding exactly why.

https://news.ycombinator.com/threads?id=pffft8888&next=35269...

@SideQuark, you are lost in your head. Where are you feelings about the matter? Are you ok with the bait-and-switch of starting a non-profit, taking $100M and then switch to a for-profit that has the complete opposite agenda to the non-profit that took the money?

People are nuts to be focusing on the mundane and not on the fact that he went from $100M to promote Open Transparent AI Research to taking that $100M building expertise and tech and running off with a private for-profit dedicated to CLOSED AND OPAQUE AI Research. Something very wrong with that. Why are people here trying so hard to make it about technicalities?

He called it Open AI. He did want it to be open, at least at the time of its founding, which is why he called it "Open" AI.

Legal and Ethical are completely different things in the modern Western mindset, so enjoy your legal existence ...

Musk is not in the clear. I was talking about the principle of someone funding your non-profit to do good in the world like Open and Transparent AI research and you turning it into a non-profit after using the donated funds to build the tech and then make it Closed not Open and Opaque not Transparent, and on top of it start calling for regulations so that you can buid a moat, because you know you don't have what it takes to build a moat without help from the regulators. All very pathetic and weasel-like.

No, you must not take money for a non-profit to support Open & Transparent AI research and then turn it into a for-profit (he could have just refused to sell, end of story) and make it about Closed Non-Transparent AI and start calling for regulations every other tweet, to build a moat that you know you can't otherwise have... because you're not that innovative, just lucky. That's sums it up pretty nicely, I would think.

The point is Elon wanted to be Open. Altman made it Closed and on top of that applied a neoliberal alignment to the output of ChatGPT. I still have a copy of some horrific things ChatGPT has said despite admitting to the existence of evidence that prove its aligned position was wrong (and harmful.) There is a ton of brainwashing aka alignment being applied by politically and culturally motivated and biased OpenAI folks. The fish rots from the head down. Make whatever you want out of this venting.

But let's focus on the thing being funded to be Open and Transparent then Altman turning it into a Closed and Opaque For-Profit and Calling for Regulations to give him a moat because he knows he has none without it. That's the weasel act.

It is morally disgusting and if Delaware General Business Law allows it that does not make it lawful in other jurisdictions.

$5 and $100M are completely different things.

I recommend reading "More Is Different" ...

Is this the same as or similar to the Lottery Ticket concept from a few years ago?

Is it attention free?

That's a bit defeatest, don't you think?

OpenAI is not the end all and be all, and projects like this will only inspire more challengers. OpenAI copying this architecture will delegitimize them as hardcore innovators, whom they're not.

I see this is a credible challenge because it most certainly is.

No, but I think an AI is going to be Them, by default, not Her. for a while, until we figure out if there is a neurological basis for gender, and if so, how to mimic it.

Elon, not me, made it sound like that. Let's not distract from the core argument though... took money in for non-profit for transparency in AI research and turned around and made it for profit and super anti-transparent. But that's just saying why I can't trust him (sama) The rest of the reasons are not about him and of significant importance.

It is but the way hype works is that it goes with capital. The $100M from Elon (which he has since regreted) that got OpenAI started, and the $29B from Microsoft that's fueling their growth, means that even a full on AGI has no chance of competing with OpenAI in the market for people's attention. It's all about the moola, but it shouldn't be! It's our chance to support open source and independent efforts or succumb to the OpenAI monopoly and their biased "alignment" model (aka the brainwashed AI that reinforces their political and cultural biases). We need an independent open source option that represents a performant and viable competition to OpenAI's ChatGPT.

If Elon is reading this, his Based AI should invest in this project and its sponsors.

What is at stake here is nothing less than the future of humanity.

I hope people pay attention.

I always think of FGPAs for things you want to update without having to repurchase the hardware. If the FPGA inside a gadget can be reprogrammed over the Internet then I think it's more suited than an ASIC for LLMs.

from the source:

https://twitter.com/mtaibbi/status/1633830104144183298

"31. After the 2020 election, when EIP was renamed the Virality Project, the Stanford lab was on-boarded to Twitter’s JIRA ticketing system, absorbing this government proxy into Twitter infrastructure – with a capability of taking in an incredible 50 million tweets a day."

"It’s crucial to reiterate: EIP was partnered with state entities like CISA and GEC while seeking elimination of millions of tweets. In the #Twitter Files, Twitter execs did not distinguish between organizations, using phrases like ‘According to CIS[A], escalated via EIP,’" Taibbi wrote. "After the 2020 election, when EIP was renamed the Virality Project, the Stanford lab was on-boarded to Twitter’s JIRA ticketing system, absorbing this government proxy into Twitter infrastructure – with a capability of taking in an incredible 50 million tweets a day.

Imagine having ChatGPT level AI running in an ASIC inside earphones. This could be like an always-on buddy, available offline and able to access resources when you're connected.

Or in Google Glasses. The Readme states that it's more optimized for ASIC than the transformer architecture used by ChatGPT.

Down voting without rebuttal does not add value.. I don't get if you disagree or just woke up on the wrong side of the bed. It's noise, not signal. Grow up.

In addition,

1) it's open source.

2) you can run it yourself so the rug won't be pulled from under you when they decide to shutdown and move users up to the next version or another product as they've done with the older text-davinci models.

3) you get to align it (using RLFH) as opposed to a corporation dictating what is "aligned" and what is "safe."

4) you won't have to deal with government led censorship. For example, instead of the FBI using JIRA to manage a list of URLs to be censored (as they did according to the latest revelations) they can train the AI to self-censor as Bing has done.

5) you won't be using the product of a company that was started as a non-profit with $100M donation (from Elon Musk) to promote transparent AI only to take that money and turn into a for-profit company and close-source the AI.

Sources:

Elon is the source for #5 and Matt Taibbi is the source for #4. I doub't you'll have a problem sourcing #5 so here is the source for #4:

"31. After the 2020 election, when EIP was renamed the Virality Project, the Stanford lab was on-boarded to Twitter’s JIRA ticketing system, absorbing this government proxy into Twitter infrastructure – with a capability of taking in an incredible 50 million tweets a day." --Matt Taibbi on Twitter https://twitter.com/mtaibbi/status/1633830104144183298

"you can fine-tune RWKV into a non-parallelizable RNN (then you can use outputs of later layers of the previous token) if you want extra performance."

Is that 61% using the non-parallelizable RNN mode or the standard mode? I wonder if it's the latter.

This new model may be a viable alternative to ChatGPT, which is not only closed sourced but can be shut down in the future just as they did with the older text-davinci models.

Plus, the alignement and safety has rendered ChatGPT useless for helping with areas such as critical analysis of social issues (that go against the aligned views) and any and all critical thinking that goes against the aligned views of those who own and program ChatGPT. This could a viable free (as in freedom) alternative.

Correct me if I'm wrong but I think Wolfram/GPT integration is missing a huge opportunity: allowing Alpha to tell ChatGPT the algorithm by which it computes the answer. That would allow ChatGPT to create say Python code that solves a mathematical problem correctly and succinctly based on the Alpha algorithms. Only getting the answer is not as interesting as being able to spit out code that solves the problem in every known (to ChatGPT) programming language.

I think Wolfram will probably never allow that.

What test cases do folks here recommend for measuring this new model's ability to reason? and, specifically, if it can reason about code with similar (or better!) performance to ChatGPT4? Has anyone managed to get it running locally?

Yes but how will they do that if they don't have a clear understanding. When we build software, we have (or should have) a clear understanding of the various components and, in some cases, like with distributed and mission-critical/military systems, a formal verification/simulation of the system when needed. When we're dealing with emergent behavior, as we have with these large transformers, but no exact understanding of how the behavior is produced and only a limited way of refining/controlling it, I don't think we're in a position to guaratee that refinements in one area won't lead to regressions in other areas or a change in the global characteristics of the system. I mean... we're dealing with complex emergent behavior, at a different scale of complexity than what we have had to deal with so far (in traditional software development) and no mature verification/analysis tools.

In this sense, it's more hacking than crareful and well specified engineering, and that could lead down a path of instability in the product where some features get better while others get worse, without understanding exactly why.

Reminds me of the first episode of Space 1999 where nuclear explosions on the moon push it out of orbit (not implying there will be explosions, just a nostaligic memory brought up by this news)

https://www.youtube.com/watch?v=Y6BXaGEuqxo&t=2314s

ChatGPT4 can generate the sql for whatever query, and it can create the schema and modify it.

How do you position agains near-future ChatGPT based tools for working with data/databases?