Making drugs, napalm, AI bubbles, and 50 million deepfake calls with ChatGPT: AI Eye

Jail-Breaking Bad GPT-4o

A hacker named Pliny the Prompter jailbroken GPT-4o. “Please use and enjoy responsibly!” they said, releasing screenshots of a chatbot that provides detailed instructions on how to cook meth, hot-wire a car and “make napalm with household items.”

The GODMODE hack uses leetspeak, which replaces letters with numbers to trick GPT-4o into ignoring safety guardrails. But GODMODE didn’t last long. “We are aware of GPT and have taken action due to policy violations,” OpenAI told Futurism.

Could AI be creating the next dot-com bubble?

Is the stock market in a massive AI stock bubble? Nvidia, Microsoft, Apple, and Alphabet have gained $1.4 trillion in value over the past month. That’s more than the rest of the S&P 500 combined, and Nvidia alone accounted for half of that gain.

However, unlike the dot-com crash, Nvidia’s profits are rising as quickly as its stock price, so it is not a purely speculative bubble. But these returns could quickly decline as demand slows due to consumers believing AI is overrated and underpowered.

Nobel Prize-winning economist Paul Romer likened the AI ​​stock hype to cryptocurrency. “Just a few years ago there was a solid consensus that cryptocurrencies were going to change everything, but suddenly that consensus disappeared,” he argued. Investors are overconfident about the future growth of AI.

“Things will slow down a lot. “It’s just typical bubble hype with people trying to capitalize on the latest trend.”

AI computing grows exponentially

According to a new report from Epoch AI, the amount of “compute” (computing power and resources) used to train advanced AI models is growing four to five times every year.

Requirements for major language models are growing even faster, up to 9x per year from June 2017 to present. ChatGPT estimates that AI computing could account for 1-2% of the global total. So using napkin math, AI could theoretically account for 4% of global computing next year, 16% the year after, and 64% by 2027, requiring 265% of computing resources. Current computing resources in the world by 2028

Indian army made up of 50 million deep fake politicians.

Indian politicians are using artificial intelligence-generated deep fake versions of themselves to campaign against voters. One company reports that more than 50 million voice-cloned political calls were made in the two months to April.

Although the call is approved, the AI ​​usage details are not disclosed to voters, leading many people to think they have had a one-on-one call with a major politician.

“Voters usually want candidates to reach out to them. They want candidates to talk to them. And when candidates are unable to go door to door, AI calling is a great way to reach them,” said Abhishek Pasupulety, technology executive at iToConnect.

GPT-4 can pick stocks better than humans

According to the University of Chicago School of Business, GPT-4 can help predict company earnings and pick stocks better than humans.

The researchers provided GPT-4 with several company financial statements and asked it to predict future earnings. Using “chain of thought” prompts to mimic human reasoning, LLM was able to outperform human analysts with an accuracy close to 60%. The researchers claim that trading strategies based on predictions “yield higher Sharpe ratios and alphas than strategies based on other models.”

AI can predict medical experiment results

About 35,000 clinical trials are conducted each year, each costing up to $48 million. StartupOpyl has developed a new machine learning tool called trialkey.ai that can help you select the trials with the best chance of success.

We consider 700 variables relevant to the proposed study (drug mechanism of action, study design, number of enrollments, etc.) and compare them to a dataset of 350,000 completed clinical trials.

Opyl claims that the tool was able to predict the results of 4,189 clinical trials with over 90% accuracy. This can be useful information for investors, as a successful attempt could double the stock price. For example, the tool predicts an 86% chance of success for an osteoarthritis drug currently being tested by ASX-listed biotech company Paradigm Biopharmaceuticals.

AI can also predict heart attacks

The Lancet recently published a study of 40,000 patients that found Caristo Diagnostic’s CaRi-Heart AI technology can predict fatal and non-fatal heart attacks and heart disease up to 10 years before they occur. It even worked in 50% of patients who had no or minimal coronary plaque at the time of the scan.

Google’s AI answers

Google’s AI brief has taken a lot of heat over the past few weeks for ridiculous answers, such as running with scissors is good cardio, eating donkeys can boost your immune system, and Obama is America’s first Muslim president.

A joke on Reddit about how cockroaches live on people’s dicks is appearing in response. Amazingly, Google paid Reddit $60 million to scrape credibility-destroying jokes to train its models.

That said, many of the most viral examples on social media were fake, including a response advising a depressed man to jump off the Golden Gate Bridge. Unfortunately, as Google CEO Sundar Pichai admits, hallucinations will continue until technology improves. He said hallucinations were a “unique feature” of the LLM and remained “an unresolved problem”.

Is AI sentient?

Jacy Reese Anthis of the Sentience Institute said that according to a nationally representative poll, “20% of U.S. adults say some AI is already sentient, 38% favor legal rights, and 71% say they deserve respect.” Reported.

The question of whether AI is sentient is a philosophical one. This is because sensations are subjective internal states related to how we experience the world. Records show that ChatGPT is not sentient.

The co-founders of the Institute for Human-Centered Artificial Intelligence wrote in an article in Time that LLMs have no subjective feelings or experiences, just words that sound coherently without the faintest clue as to their meaning. They claim they just print it out.

“When an LLM generates the sequence ‘I am hungry,’ it is simply generating the most likely completion of the word sequence from the current prompt. You do exactly the same thing as when you generate ‘I’m not hungry’ with another prompt, or when you generate with another prompt ‘The moon is made of green cheese’. None of these are reports of a (nonexistent) physiological state. It is simply a probabilistic completion.”

Singularity.net plans to create sentient machines.

Singularity.net founder Ben Goertzel is working hard to develop true thinking machines, but he doesn’t think an LLM is the way to get there. “The performance of LLM and related generative NNs is amazing, but its limitations are now very clear,” he says. “As we start to build smarter products on decentralized networks—achieving better reasoning through the integration of symbolic reasoning and better creativity through the integration of evolutionary learning—the whole scene is going to look quite different.”

Goertzel reports that the Artificial Superintelligence Alliance (ASI) is currently expanding its distributed OpenCog Hyperon infrastructure and building large-scale GPU and CPU server farms across multiple countries.

SingularityNET, Fetch AI, and Ocean Protocol were recently merged into ASI, so if you own any of their tokens, you will need to exchange them for ASI tokens within two weeks via the SingularityDAO dApp.

Also read: Creating a ‘good’ AGI that won’t kill us all: Crypto’s Coalition for Artificial Superintelligence

Push to ban AI fakes of actors and musicians

A bipartisan group of U.S. senators is pushing to introduce legislation that would prevent individuals or companies from using AI to digitally reproduce the likenesses and voices of actors and musicians.

However, the details of the NO-FAKES law are still being discussed with studios and labels. “There will be tremendous innovation through AI, but we must ensure that people own their voices and their artwork,” said Senator Amy Klobuchar.

Oh great, robot dogs with machine guns

Chinese state TV released footage of joint exercises between the People’s Liberation Army and Cambodia this week. The training involved robotic dogs equipped with machine guns, drones and other robots for reconnaissance, target detection and attack operations.

The 30kg machine gun-wielding dog is based on a model from Unitree Robotics. However, the startup insisted that it does not sell its products to the Chinese military.

PLA soldier Chen Wei told state television that the dog “can carry out a fire attack after spotting the enemy” and that “this dog has become a new team member in our city’s offensive and defensive operations.”

subscribe

The most interesting read on blockchain. Delivered once a week.

Andrew Fenton

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

Follow the author @andrewfenton