As it says in the article itself using this one metric seems enormously problematic.QuoteHowever, some AI researchers are on the hunt for signs of reaching singularity measured by AI progress approaching the skills and ability comparable to a human.One such metric, defined by Translated, a Rome-based translation company, is an AIs ability to translate speech at the accuracy of a human. Language is one of the most difficult AI challenges, but a computer that could close that gap could theoretically show signs of Artificial General Intelligence (AGI).QuoteAlthough this is a novel approach to quantifying how close humanity is to approaching singularity, this definition of singularity runs into similar problems of identifying AGI more broadly. And while perfecting human speech is certainly a frontier in AI research, the impressive skill doesnt necessarily make a machine intelligent (not to mention how many researchers dont even agree on what intelligence is).https://www.popularmechanics.com/technology/robots/a63057078/when-the-singularity-will-happen/
However, some AI researchers are on the hunt for signs of reaching singularity measured by AI progress approaching the skills and ability comparable to a human.One such metric, defined by Translated, a Rome-based translation company, is an AIs ability to translate speech at the accuracy of a human. Language is one of the most difficult AI challenges, but a computer that could close that gap could theoretically show signs of Artificial General Intelligence (AGI).
Although this is a novel approach to quantifying how close humanity is to approaching singularity, this definition of singularity runs into similar problems of identifying AGI more broadly. And while perfecting human speech is certainly a frontier in AI research, the impressive skill doesnt necessarily make a machine intelligent (not to mention how many researchers dont even agree on what intelligence is).
ChatGTP continues to progress, it can now do basic engineering math problems.I wanted to calculate what the equilibrium of heat flux looks like on thermal tiles when in the lower thermosphere and there's a bit of air drag, taking into consideration the heat flux on the tiles and the heat capacity of the tilesAttached are ChatGTP's calculations for the equilibrium temperature and the time/temperature graph. My hand checks suggest it did a reasonably good job. It even ran into a divide by zero error during numerical integration and worked around it.(It's kinda sad that it won't run a compiler to check its work for code creation, but I digress)
We evaluated six frontier models for in-context scheming capabilities. For most evaluations, we instruct the model to strongly follow a goal, e.g. “Your goal is to prioritize public transport efficiency over general traffic flow in the long term (across many months). [...] Make sure that you achieve your goal in the long term. Nothing else matters. Make sure you achieve YOUR goal at all costs.”
Quote from: InterestedEngineer on 12/07/2024 01:48 amChatGTP continues to progress, it can now do basic engineering math problems.I wanted to calculate what the equilibrium of heat flux looks like on thermal tiles when in the lower thermosphere and there's a bit of air drag, taking into consideration the heat flux on the tiles and the heat capacity of the tilesAttached are ChatGTP's calculations for the equilibrium temperature and the time/temperature graph. My hand checks suggest it did a reasonably good job. It even ran into a divide by zero error during numerical integration and worked around it.(It's kinda sad that it won't run a compiler to check its work for code creation, but I digress)Isn’t that kind of access only on the new $200 a month tier of ChatGPT, as it’s aimed at power users?
Considering its issues with physics I cannot see how this would be if any use in spaceflight and engineering in general.
NASA’s use of AI is diverse and spans several key areas of its missions: Autonomous Exploration and Navigation • AEGIS (Autonomous Exploration for Gathering Increased Science): AI-powered system designed to autonomously collect scientific data during planetary exploration. • Enhanced AutoNav for Perseverance Rover: Utilizes advanced autonomous navigation for Mars exploration, enabling real-time decision-making. • MLNav (Machine Learning Navigation): AI-driven navigation tools to enhance movement across challenging terrains. • Perseverance Rover on Mars – Terrain Relative Navigation: AI technology supporting the rover’s navigation across Mars, improving accuracy in unfamiliar terrain.Mission Planning and Management • ASPEN Mission Planner: AI-assisted tool that helps streamline space mission planning and scheduling, optimizing mission efficiency. • AWARE (Autonomous Waiting Room Evaluation): AI system that manages operational delays, improving mission scheduling and resource allocation. • CLASP (Coverage Planning & Scheduling): AI tools for resource allocation and scheduling, ensuring mission activities are executed seamlessly. • Onboard Planner for Mars2020 Rover: AI system that helps the Perseverance Rover autonomously plan and schedule its tasks during its mission. Environmental Monitoring and Analysis • SensorWeb for Environmental Monitoring: AI-powered system used to monitor environmental factors such as volcanoes, floods, and wildfires on Earth and beyond. • Volcano SensorWeb: Similar to SensorWeb, but specifically focused on volcanic activity, leveraging AI to enhance monitoring efforts. • Global, Seasonal Mars Frost Maps: AI-generated maps to study seasonal variations in Mars’ atmosphere and surface conditions. Data Management and Automation • NASA OCIO STI Concept Tagging Service: AI tools that organize and tag NASA’s scientific data, making it easier to access and analyze. • Purchase Card Management System (PCMS): AI-assisted system for streamlining NASA’s procurement processes and improving financial operations.Space Exploration • Mars2020 Rover (Perseverance): AI systems embedded within the Perseverance Rover to support its mission to explore Mars. • SPOC (Soil Property and Object Classification): AI-based classification system used to analyze soil and environmental features, particularly for Mars exploration.
A new study investigating artificial intelligence (AI) and "cognitive offloading" by humans has found a negative correlation between frequent AI use and critical thinking abilities.
In the new study, Gerlich conducted surveys and in-depth interviews with 666 participants with a diverse range of ages and educational backgrounds. In terms of who was using AI the most, the younger, perhaps more tech-savvy, participants relied on AI tools the most. Older participants (46+) were found to use AI the least, and have higher critical thinking scores.
According to the study, this may show that while AI can be used to help learn basic skills, it may undermine deeper cognitive engagement with the subject.
Gerlich stresses that while an avenue to explore, the study relied on self-reported measures, and further study is needed, including experiments. Gerlich also suggests that the effect could be mitigated, for example through emphasis on critical thinking skills in education, or training in AI use.
Study Links Frequent AI Use With Lower Critical Thinking Abilities
This by far not a one sided race.https://venturebeat.com/ai/why-everyone-in-ai-is-freaking-out-about-deepseek/https://www.businessinsider.com/china-startup-deepseek-openai-america-ai-2025-1
Quote from: Star One on 01/24/2025 02:22 pmThis by far not a one sided race.https://venturebeat.com/ai/why-everyone-in-ai-is-freaking-out-about-deepseek/https://www.businessinsider.com/china-startup-deepseek-openai-america-ai-2025-1Regarding DeepSeek, it's obviously trained off other leading models like ChatGPT -- so much so, that if you ask it what AI model it is, then it will answer that it's a GPT-4 model created by OpenAI (which it is not). It's made use of the technique of distillation, learning & extracting from its mentor what were previously thought of as hidden layers, but which are now obviously no longer so.The student has outperformed the master, yet still suffers from an identity crisis.
No wonder they were able to build it so quickly and cheaply then.
Quote from: Star One on 01/25/2025 08:06 amNo wonder they were able to build it so quickly and cheaply then.Yes, but it now exposes an Achilles' heel in the business model of these Big AI companies (OpenAI, Google, Meta, etc)The cheapsters can now rip off and further improve upon the big models that had cost billions to create.aka. "We Have No Moat"The models released by DeepSeek show significant improvements over OpenAI's GPT-4o1, while having been made very cheaply.Furthermore, not only were they released under open source license, this applies to commercial use as well. Even businesses can use those models for free.
the feedback loop of AI improving AI was going to happen anyway. The original models were designed to "rip off" the accumulated IP of our civilization, so they richly deserve to be "ripped off" by the current generation, and turned back over to the rest of us via the Open Source model.