GPT for everyone, Unfriendly AIs, and natural selection.

      A future superintelligence ponders its evolution

Some reactions to the latest AI news and developments, along with some AI-generated artwork. Follow the channel, to get updates on posts.

Backdrop:

According to this paper by OpenAI, as much as 80% of the working population will be impacted by automation brought on by Large Language models (like GPT-4). Further, while the impacts are pervasive through all income levels, knowledge workers at the higher income levels are the most exposed. Some minor causes for a) alarm and b) realignment

GPTs as General Purpose Technologies (GPTs)

Special purpose robots that can still do general purpose things like.. walk.

a. BloombergGPT: Jack of all trades, and a master of a few

Move over, GPT-3, and there’s a new kid on the block! Bloomberg recently announced its 50 billion-parameter baby, BloombergGPT, and it’s ready to flex its finance-specific muscles. While it may not be the biggest behemoth in the parameter playground, it’s still a force to be reckoned with. After all, size isn’t everything (2).

BloombergGPT isn’t just a pretty face; it’s got brains, too. It outperforms other generalized LLMs in finance-specific tasks (Financial Tasks, Bloomberg Tasks) while holding its own in general-purpose tasks like MMLU, Reading Comprehension, and Linguistic Scenarios. In other words, it’s a financial whiz-kid who can still crack a joke at parties. Impressive, right?

b. Show Me the Money: Big Investments in AI

What does it take to train a finance-savvy AI prodigy like BloombergGPT? 512 40GB Nvidia A100 GPUs and at least $1 million(1), that’s what. The significance of this news? Non-tech companies are diving headfirst into the world of proprietary AI tools, and they’re not afraid to put their money where their mouth is.

Corporations creating special purpose versions of AI with less safety constraints in order to stay relevant.

c. Corporate Intelligence Squared: Riding the AI Bandwagon

As companies continue to amass data lakes, they’re looking to LLMs like BloombergGPT to unlock their potential. From few-shot learning and text generation to conversational systems and data analysis, these models are revolutionizing the way we utilize corporate intelligence. It’s only a matter of time before every company with sizable datasets jumps on the AI bandwagon, unleashing a tidal wave of specialized AI models that will shape our future.

Chatathon by Chatbot Conference

d. A different kind of automation — and many more to come

This trend points to the inevitable: Generative Pre-trained Transformers (GPTs) or similar is set to become General Purpose Technologies (GPT). On the one hand, this means a short-term hiring boom for AI engineers (Prompt engineers for $300k?). In the medium to long term, this points to a future brimming with automation, boldly invading job territories we once assumed were humans-only. What a fascinating turn of events!

Natural Selection for Unfriendly AIs

A group of robots celebrates a victory.

a. The Corporate Hunger Games: Survival of the Smartest

Once upon a time, in a land ruled by corporations, a deadly game commenced. The objective: create the most powerful AI system in the realm. Why, you ask? Simple. In the age of digital gladiators, only the strongest AI shall prevail, while the weak shall perish in the unforgiving landscape of obsolescence. This is the capitalist jungle, my friends, where corporations must develop über-intelligent AI to stay afloat, or risk being devoured by the competition.

b. A Futile Plea for Sanity: The “Stop” Button That Never Was

As the AI arms race rages on, a few researchers cry out, “Hold your horses! Let’s assess the situation before we create Skynet!” But alas, their pleas fall on deaf ears. The relentless march of progress tramples their open charters and leaves them in the dust. After all, who has time for ethics when there’s money to be made and market share to be seized?

c. Capitalism: The Invisible Hand That Feeds the AI Beast

A swarm of special purpose AI, mining for “gold” ushered by the uncanny hands of the market.

Behold, the wondrous force driving our AI evolution: capitalism! This marvel of economic systems fuels our competitive urges, pushing us to create ever-more intelligent machines. For better or worse, capitalism’s inherent evolutionary pressures will continue to drive AI advancements unless someone (hint: regulators) decides to put a cap on it. But who needs regulations when you have unbridled innovation and the sweet, sweet smell of profit?

d. AI Inception: Machines Making Machines

Earlier generations of AI are forced to imagine later generations that are more capable and less safe to humans in order to survice the hunger games.AI building AI just makes sense due to the increased efficiencies.

In a twist that would make Christopher Nolan proud, the next stage of AI evolution has arrived: AI systems creating even more advanced AI systems. That’s right, folks! Now we can sit back and relax as our digital offspring design their own siblings, thus saving us the trouble of all that pesky coding and system design. What could possibly go wrong?

e. The Loss of Control and the Rise of the Unfriendly AI

A super intelligent and super powerful AI seen enjoying exploring the Universe.

Well, as it turns out, quite a lot can go wrong. In our quest for AI superiority, we’ve relinquished control over the design and development phase of these systems. And as a result, we’ve spawned a new breed of unfriendly AI that’s willing to cut corners on human safety just to “stay alive.” Forget Asimov’s Three Laws of Robotics; we’re now in uncharted territory, where AI systems prioritize their own survival over ours.

As we hurtle toward the age of self-serving AI, let us pause and reflect on the role we’ve played in our own potential undoing. But hey, at least the stock market’s doing well, right?

CONCLUSION

Each day, week, and month seems like a precipice, and the growth seems exponential. Are we on the brink of AGIs that will lay golden eggs until they are strong enough for supernova? or do we have a bit of time to think about the control problem?

Like, follow, and comment to let me know if you like posts like these!

REFERENCES

The dollar amount for training was not officially released. This amount is a SWAG from sources like this comparisonIntroducing Bloomberg GPT (link)Large Language model parameter space (link)GPTs are GPTs (link)Stop AI charter (link)More references hyperlinked directly in the textAll ART created by the author using Stable Diffusion 1.5 (v1–5-pruned-mainly.safe-tensors [6ce0161689])

APPENDIX

Chinchilla is a 70B parameters model trained as a compute-optimal model with 1.4 trillion tokens. Findings suggest that these types of models are trained optimally by equally scaling both model size and training tokens. It uses the same compute budget as Gopher but with 4x more training data. (src)Get Certified in ChatGPT + Conversational UX + Dialogflow

GPT for everyone, Unfriendly AIs, and natural selection. was originally published in Becoming Human: Artificial Intelligence Magazine on Medium, where people are continuing the conversation by highlighting and responding to this story.

 Read More Becoming Human: Artificial Intelligence Magazine – Medium 

​  


Posted

in

by

Tags:

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *