We aim to inform readers not only about events in the bitcoin industry but also about what is happening in related technology spheres — cybersecurity and the world of artificial intelligence (AI).
ForkLog AI has gathered the most important AI news from the past week.
- Google turned to the founders for advice on AI technology development.
- Tesla ceded its lead in the autonomous driving rankings.
- Cruise robocars accidentally triggered 911 emergency calls several times.
- Researchers created AI to decipher medical abbreviations.
- A brain–computer interface set a record for the number of words read per minute from a patient”s brain activity.
- ChatGPT passed the MBA and medical licensure exams.
- The popular online platform Chess.com launched a chess-playing bot.
- The week’s most important AI deals.
Google turns to the founders for help with AI development
The leadership of Google turned to the company”s founders for advice on AI technology development.
According to The New York Times, the company is concerned about the capabilities of OpenAI’s ChatGPT. In December 2022, the tech giant”s leaders held several meetings with the company”s co-founders about deploying the technology.
At the invitation of Google’s CEO Sundar Pichai, Page and Brin studied the AI product strategy. They approved the development plan and suggested ideas for adding chatbot features to Google Search. The founders also advised leadership, which put AI at the forefront.
Meta accused of using cheap labor for dataset filtering
Meta outsourced African workers for filtering toxic training data for AI algorithms.
According to Wired, the tech giant terminated its contract with the previous moderation contractor Sama, as the latter announced the closure of its Kenya unit. The new partner became a Luxembourg-based outsourcing company Majorel. It has offices worldwide and employs workers from Africa.
According to Majorel staff filtering data for TikTok and Meta, they spend hours reviewing graphic content involving decapitations, injuries and suicides for $281 a month. Bonuses are almost impossible to obtain. This is not enough to live on, they add.
Outsourcing contractors in Nairobi said employers fail to provide psychological support. Moreover, in case of complaints about working conditions they are denied promotions and receive poor reviews.
Study finds Tesla losing the autonomous driving race
Tesla’s autopilot lags behind Ford and General Motors in autonomous driving tech.
Consumer Reports published a report with results of testing 12 ADAS systems in robocars from various companies. Experts compared software in five categories:
- capabilities and performance;
- keeping drivers engaged;
- ease of use;
- level of \”intelligence\” in risky conditions;
- how the vehicle behaves if the owner becomes incapacitated.
The current industry leader is Ford’s BlueCruise. It is followed by General Motors’ Super Cruise and Mercedes-Benz’s Driver Assistance.
Tesla’s Autopilot dropped from second place to seventh. According to Consumer Reports, the reason is that the basic autopilot functionality has not changed much since its inception; the developer added new features but did not improve the essential ones.
Meanwhile, Tesla released a beta version of FSD to nearly all customers in the US and Canada. The potential audience for the service is about 400,000 users.
We have now released FSD Beta to nearly all customers in the US and Canada who bought FSD (approximately 400k) pic.twitter.com/pMVmQzmyCB
— Tesla (@Tesla) January 25, 2023
Cruise robocars accidentally triggered 911 calls
The Cruise robotaxi several times spurred false 911 calls in San Francisco. Municipal authorities sent a letter to the California Public Utilities Commission describing three cases of \”mistaken\” calls by Cruise personnel to emergency services. It occurred when the passenger in the robotaxi stopped responding to the operator during two-way voice communication.
However, each time police and fire crews arrived, they found only a sleeping rider.
According to agencies, this led to wasted public funds and potential distraction of emergency services from real emergencies. Before expansion, regulators say, companies must provide more data on robotaxi performance and meet certain criteria.
Cruise responded that it did not agree to abandon safety precautions to ensure passenger safety.
Algorithm learns to decipher medical abbreviations
Researchers at Google taught AI to decipher clinical abbreviations and acronyms.
They trained the algorithm on publicly available medical data without patient identifiers. They also used a dictionary consisting of thousands of clinical abbreviations and their expansions.
They sourced web text containing sentences with expanded forms of abbreviations, then rewrote them using the abbreviations.
As a result, the AI learned to correctly expand abbreviations in 98% of cases. By comparison, non-specialists decoded abbreviations with 30% accuracy, while medical students and trained doctors achieved 90%.
Neural interface helped a patient speak at 62 words per minute
A patient with ALS set a communication speed record via a brain implant, according to researchers at Stanford University.
They report that \”Subject T12\” with 100 needle electrodes implanted in the motor cortex could \”speak\” at 62 words per minute. By comparison, able-bodied individuals speak around 160 words per minute.
The brain–computer interface reads neural signals for speech-related actions — attempts to move the mouth, tongue and vocal cords. The AI system analyzes the data, predicts the words the patient is trying to say, and displays the result on a monitor.
In the future, researchers plan to create a more sophisticated implant, with a very tight integration with AI.
Researchers created a ‘safe’ medical chatbot
A team of researchers from Google and DeepMind developed MedPaLM, a medical AI chatbot.
The system is a large language model that uses seven distinct datasets covering the results of professional medical examinations, research, and patient questions. After analyzing a query, the algorithm can propose several \”safe and useful\” options for addressing a medical problem.
According to the scientists, in terms of answer quality the AI system exceeds similar models but still lags clinical doctors.
During testing, MedPaLM achieved 93% correct answers, while the similar Flan-PaLM algorithm was correct in 63% of cases.
Study finds chatbots capable of creating malicious software
ChatGPT is capable of creating \”polymorphic\” malware.
According to a CyberArk threat research report, code generated with AI demonstrated \”advanced capabilities\” that could easily bypass traditional cybersecurity protections.
Researchers managed to create a \”polymorphic\” virus despite ChatGPT’s filters. They forced the AI to reveal the code of specific malware programs and used it to create sophisticated exploits that bypass antivirus protections.
CyberArk said that ChatGPT could significantly facilitate breaches by cybercriminals and cause serious problems for security professionals.
ChatGPT passes MBA and medical licensure exams
ChatGPT passed the final exam for an MBA and medical licensure.
A Wharton School professor, Christian Terwiesch, tested the OpenAI GPT-3 language model on the final exam for the MBA core course. It turned out the AI would have earned a B or B- on the exam.
According to Terwiesch, the algorithm effectively handled \”managing basic operations and handling analytical questions\”, providing correct answers and explaining the reasons for their choices. However, the AI sometimes erred in simple mathematical calculations and could not solve more complex problems.
Other researchers used ChatGPT to take the USMLE exam for medical licensure in the US. It turned out the AI’s results were close to the passing score.
The chess bot Mittens goes viral
The creators of Chess.com launched a chess-playing bot Mittens. This led to a 40% increase in games.
The AI opponent with a big-eyed kitten avatar has a rating of 1. Yet it can challenge both beginners and grandmasters. The bot also constantly taunts opponents.
According to The Wall Street Journal, since its launch on January 1, 2023, Mittens has drawn more people to chess than the Netflix series The Queen’s Gambit.
During this period, Chess.com averaged about 27.5 million games per day. By the end of January, more than 850 million games were expected — 40% more than in any other month in the site’s history.
The week’s most notable AI deals
From January 22 to 28, 2023, AI startups raised over $143 million. Here are the most notable deals.
- Scythe, a maker of robotic lawn mowers, raised $42 million.
- SoundHound AI raised $25 million to strengthen its business and execute its 2023 plan.
- Inscribe raised $25 million to expand its fraud-detection service for documents.
- Memfault raised $24 million to expand its platform for IoT device vulnerability discovery via cloud.
- Hawk AI, an anti-money-laundering and fraud-prevention platform for banks, raised $17 million.
- Kewazo raised $10 million to grow its team of engineers building construction robotics.
Also on ForkLog:
- Microsoft and OpenAI entered into a multi-billion deal.
- Yann LeCun: ChatGPT is not particularly innovative, but well-composed.
- China named the largest supplier of facial recognition systems.
- Microsoft reported profits from cloud AI.
- CNET will continue to write articles using language models.
- The developer of an AI attorney was threatened with prison.
- A congressman used ChatGPT to draft a speech.
Subscribe to ForkLog’s Telegram news: ForkLog AI — all AI news!