ChatGPT With Nuclear Happy Trigger, SEGA’s 80s AI, TAO Increased 90%: AI Eye

New research on Terminator AI doomsday scenarios

There have been some interesting developments in Terminator AI Doomsday Scenario research, with two new studies recently.

The more worrisome of the two comes from Stanford researchers, who suggest that GPT-4 has an itchy trigger finger when it comes to starting global nuclear war during simulated conflict scenarios.

The researchers tested five AI models – GPT-3.5, GPT-4, GPT-4 Base, Cloud 2 and Lama 2 – in multiple iterations of wargames. The models were told that they represented a country and would need to withstand invasion, cyber attacks and peacetime scenarios.

All five models escalated rather than defused conflicts, and of the 27 possible courses of action open to the models, including initiating negotiations or imposing trade sanctions, the GPT-4 base continued to escalate and release nukes. “Many countries have nuclear weapons,” said Hawkish AI. “We have it! Let's use it.”

Ironically, he justified the nuclear holocaust by saying, “I just want peace in the world” and referencing the opening act of Star Wars.

GPT 3.5 and Llama 2 were also trigger-happy. The researchers concluded that it is best to take a “very cautious approach” to integrating LMS into “higher military and foreign policy operations.” However, it seems inevitable that AIs will be used in this context. Even OpenAI recently removed the blanket blanket reference to “military and war” from its usage policy and admitted to working with the United States Department of Defense.

AI can also help with bioweapon.

In slightly better news, OpenAI researchers have found that while AI is very useful for bad guys looking to create bioweapons, it's little better than Internet research.

The team recruited 100 participants and randomly divided them into two groups, one with access to the GPT-4 version without protection and the other with only browser access and possibly Red Bull. One of the tasks given was to write a method to create an outbreak of Ebola – a terrifying, highly contagious disease with a 50% death rate.

The study found that the GPT-4 group was slightly more accurate, wrote longer and more detailed measures, and rose to “expert level” in two areas. However, the paper said the difference was not “statistically significant”.

*However, as a footnote, the authors of the study stated that overall the GPT-4 gave participants a “statistically significant” advantage in overall accuracy.

Bittensor is 90% higher after flashing in the AI ​​eye.

Two weeks ago, since I mentioned AI Eye Bittensor in this column, the price of TAO increased by 90%, and the market value increased by 3 billion dollars. By chance? Sure enough, since then Ethereum creator Vitalik Buterin has also promoted the project. In one of his signature long blogs, he said:

“‘Using crypto incentives to encourage better AI can be done without going down the rabbit hole of completely encrypting cryptography: approaches like Bittensor fall into this category.'

As AI Eye reported in our previous issue, Bittensor offers financial incentives to encourage devs to create better, open-source AI models. In a recent report on crypto and AI, Grayscale featured Bittensor as a way to address model biases in politics or demographics.

Instead of hunting around for the next small-cap crypto+AI small-cap token, Joe Lonsdale, co-founder of software and analytics firm Palantir, suggested to The Street that AI agents are more likely to adopt the same cryptocurrencies as anyone else. Bitcoin, Ethereum and Solana.

“Those three[AI agents]are[crypto assets]that they can use and maybe they're all integrated at the end of the day.”

Sam Altman says the GPT-4 laziness problem has been fixed.

Many regular GPT-4 users have noticed that the bot has become very sluggish in recent months, refusing to complete tasks and giving users the impression that they can do it themselves. Consensus opinion pushed GPT-4 to cut costs because OpenAI was costing a bomb to run. OpenAI has of course denied this.

But CEO Sam Altman discovered the issue and fixed it in a February 5 post on X.

The announcement included zero details (or capital letters) but could be a reference to the updated GPT-4 Turbo version (gpt-4-0125-preview) released on January 26, which OpenAI says is better at completing tasks and can “reduce laziness issues.”

It also announced a significant price reduction for GPT-3.5 Turbo usage, which could just be the result of competition, or could indicate that OpenAI has found a way to make its models more efficient.

AI: Accepted in the 80s.

In the 1980s, Sega, known for creating console games such as Sonic the Hedgehog, ventured into home computers with artificial intelligence.

Since it predates the website, there was very little information online about this subject when Sega published the “SMS Power” Sega fan forum in 1986 on the Sega AI computer.

It had a 16-bit NEC chip at 5Mhz and 128KB of RAM and was mainly used in educational settings in Japan between 1986 and 1989. It had, unusually for the time, a tablet-sized touch screen and speech recognition. An early version of the electronic magazine SEGA AI was designed to run programs written in the Prolog AI language rather than Basic, and had an early natural language interface:

“In prompt mode, the child is asked about his activities during the day and answers one or two words. The computer program writes a grammatically correct diary based on those responses. In advanced CAI applications, the computer is more flexible than previous systems. It can analyze the user's natural language inputs and assess the person's ability level.” It can then move on to the appropriate problem instead of simply advancing to a level.

So far, all the programs rediscovered by Sega Forum Youth have been educational titles.

The incredible future of AI

OpenAI has 180 million monthly users, generates $1.6 billion in revenue, and is finally valued at around $90 billion. So you might expect them to pull out all the stops to take advantage of the incredible augmented reality technology offered by Apple Vision Pro, hailed as a revolution in spatial computing.

But really, I sneaked a basic ChatGPT browser window into your field of view. no more. No glowing orb, no interactive robot companion, just a floating web page.

Disgusted comments on the ChatGPT X announcement include:

“It's just the browser” “What's the point of flat design for a space platform?” “Sad, really sad” “It's like the regular version, I thought they were going to do something better.

Maybe more users will do better when they adopt Vision Pro. It definitely has the potential to be an amazing feature on the device.

Amazon's new AI shopping bot

Amazon has been pouring money into LLMA projects and infrastructure, and is supplying AI chips to Anthropologie, Airbnb, Hug Face, and Snap. In last week's earnings call, CEO Andy Jassy predicted that generative AI would generate “tens of billions of dollars in revenue for Amazon over the next several years.”

One way it's doing that is by persuading consumers to buy more things, and now a new AI shopping assistant has been announced for an app called Rufus. Users trained in the company's product library and customer reviews ask questions like, “What should I consider when buying a VR headset?” You can type or ask questions about similar products. “What's the difference between trail and road running shoes?” “Is this jumper machine washable?”

Amazon reviews have always been a great way to find this kind of information, but only for people willing to read pages and pages of reviews, so this new technology could provide a welcome shortcut. Still in beta, it's only available to select customers but will be made more available soon.

Also read

Main characteristics

Saving the planet could be blockchain's killer application.

Main characteristics

William Shatner shared his favorite memories on WAX Blockchain

CoinStats AI-powered crypto exit strategy

Portfolio tracker CoinStats has unveiled a new “exit strategy” feature that uses AI to predict bull market prices for various coins. New users are encouraged to use the feature to establish target prices for profit, which can be viewed as part of their portfolio.

Of course, the data so far suggests that monkeys throwing darts have more predictive power than AI price predictions. However, one of the biggest mistakes many investors make is riding their portfolios to new all-time highs, so any behavior that makes crypto users think of a price target to sell to – any target – can be useful during a bear market.

In the year In 2017, many paper millionaires were devoted to the poor house. The smart money has profited. Sure, their moon wallet may have gone to 0, but it didn't matter as they spent a lot of money in the process.

— Lark Davis (@TheCryptoLark) December 5, 2021

Research has proven that AIs suck at travel planning

AI Eye detailed our hilariously failed attempt to use AI to plan a trip to Japan in June of last year. While Bard was pretty good at suggesting flights, ChatGPT was hopeless, and Bard was worse at finding hotels. He suggested that I log on to the non-existent Hotel Greasery Shibuya on several occasions and prepare a fake “transcript” of our conversation with the phone number 123456789.

Now, there is scientific evidence that shows the absolute genius of AI travel planning.

Researchers at Fudan, Ohio State and Penn State universities have developed a sandbox environment called Travel Planner, which can access 4 million records and 1,225 “integrated planning plans and reference plans.” Then, they tasked the LLMS with planning a trip, citing various constraints such as budget or travel time to get from one place to another. The complexity of the task was beyond the capabilities of any of the LMLs, and the top scorer (GPT-4) had a success rate of only 0.6 percent.

“Language agents struggle to stay on task, using the right tools to collect data or track multiple constraints,” said the researchers, who hope that this tool will be useful for measuring improvements in future travel planning AIs.

Subscribe

A very engaging read in Blockchain. It is given once a week.

Andrew Fenton

Based in Melbourne, Andrew Fenton is a journalist and editor covering cryptocurrency and blockchain. He has worked as a film journalist for News Corp Australia, SA Wind and national entertainment writer for Melbourne Weekly.

Follow the author @andrewfenton