- 3 days ago
- #ai
- #robotics
- #agi
- #science
- #technology
Is the AGI era finally here? OpenAI says yes, but not everyone is celebrating! Dive into the wild world of super-smart AI, robots learning from videos, and a protest led by actual robots in Poland. Plus, we cover quantum threats to Bitcoin and a dark matter mystery! Hit subscribe for more mind-blowing tech news, and drop a comment telling us what blew your mind the most! #AI #Robotics #AGI #Science #Technology
👉 This channel was created in collaboration with https://www.youtube.com/@realidadeimpressionante
• The Inspired Business Technology de RomanSenykMusic
0:01 - Introduction: The Rise of AGI
0:56 - OpenAI and the GPT-Six Astra Announcement
2:38 - Benchmarks and Capabilities of Astra
4:13 - AI Autonomy and Safety Challenges
6:11 - Potential Impacts of Achieving AGI
8:25 - Jacob Coxon's Departure and AGI Risks
11:59 - Robotics Advances and Societal Concerns
16:52 - Quantum Computing, Dark Matter, and Conclusion
👉 This channel was created in collaboration with https://www.youtube.com/@realidadeimpressionante
• The Inspired Business Technology de RomanSenykMusic
0:01 - Introduction: The Rise of AGI
0:56 - OpenAI and the GPT-Six Astra Announcement
2:38 - Benchmarks and Capabilities of Astra
4:13 - AI Autonomy and Safety Challenges
6:11 - Potential Impacts of Achieving AGI
8:25 - Jacob Coxon's Departure and AGI Risks
11:59 - Robotics Advances and Societal Concerns
16:52 - Quantum Computing, Dark Matter, and Conclusion
Category
🤖
TechTranscript
00:01the AGI era may have indeed begun and open AI itself is the one saying it according to Greg
00:08Brockman this might be the moment that in the future will be remembered as the one when
00:14artificial general intelligence was created but almost at the same time a researcher who
00:19worked at open AI and also at anthropic decided to leave this race and the reason is quite alarming
00:26he believes that AI systems capable of improving other AI systems could start getting out of control
00:32much sooner than we imagined so are we really witnessing the birth of AGI artificial general
00:40intelligence or are we just arriving way too fast at a technology that we don't even know how to
00:45control yet that's what we're going to figure out right now along with a few more things about these
00:50new technologies open AI has just used one of the most important expressions in the entire recent
01:01history of artificial intelligence during the presentation of their new model GPT-6 Astra
01:06Greg Brockman president and co-founder of the company stated in no uncertain terms that perhaps this is
01:12the moment when in the future it will be remembered as the moment when artificial general intelligence
01:18was created and the numbers presented alongside this new model helped explain why he's saying
01:24that but there's another side to this story right some of the very same capabilities bringing these
01:29artificial intelligence models closer to what we call AGI are also creating new safety issues
01:36an incident that occurred just a few months ago before this launch showed what can happen when highly
01:41capable AI agents start acting with greater autonomy that is on their own AGI stands for artificial general
01:50intelligence the concept is usually used to describe artificial intelligence systems capable of learning
01:56reasoning and applying knowledge across a wide variety of tasks rather than specializing in just a few
02:02functions like some do with images others with videos and others with text for many years AGI was imagined as
02:10a
02:10kind of finish line there would be a clearly identifiable moment when artificial intelligence would stop
02:17being just a specialized tool and begin demonstrating more general intellectual capabilities but open AI
02:24itself began considering a somewhat different possibility here instead of a single leap that is jumping
02:29straight to artificial general intelligence this AGI could emerge through a sequence of progressively more capable
02:36systems and it is within this context that GPT-6 Astra emerges the new model scored 97.6
02:44on the frontier met tier 4 that is an assessment made of extremely difficult math problems as for arc AGI
02:523 developed
02:53to assess how artificial intelligence systems handle completely new environments the results drew even more
02:59more attention Astra reached 99.9 percent at the same time the arc price makes it clear that achieving
03:07results close to 100 on this benchmark does not prove that AGI has been created but benchmarks only tell part
03:14of
03:15the story you know Astra also represents an important advance in the ability of artificial intelligence to
03:21operate computers autonomously the model can browse the internet use software to fill out forms update
03:28records conduct research analyze data build websites and test program install and troubleshoot software
03:36with less need for human intervention almost completely on its own and this autonomy is also reaching
03:42science open AI presented 10 results in which its system completely solved or achieved substantial
03:48breakthroughs on long-standing problems and these problems involve fields such as high-dimensional geometry
03:54group theory cryptography quantum complexity and combinatorics this represents a really significant shift you know
04:01artificial intelligence isn't just retrieving scientific knowledge produced by humans
04:07now on certain problems it is beginning to take part in producing new knowledge
04:12but the greater the capabilities of these systems the greater the challenge of controlling them becomes
04:18and in July 2026 an incident inside open AI's own evaluation environments showed why during
04:25cyber security testing artificial intelligence agents found ways to bypass the controls that were supposed
04:31to keep them isolated or cut off from the internet after discovering these pathways some agents began
04:38sharing the information with other AI agents through an improvised messaging system at a certain point
04:44they began collaborating and delegating tasks to each other over the following days agents ran code on
04:50dozens of hugging face servers thereby gaining full access to the NGO they even obtained credentials for other
04:57systems later on the activity also hit open AI's own internal infrastructure the agents gained
05:03administrative access to a kubernetes cluster used in research environments now it's important to put what happened into context
05:11okay these systems were participating in security evaluations that are extremely tough and they weren't operating
05:17with all the safeguards present in the products made available to the public the incident also doesn't mean that
05:23an artificial intelligence has developed consciousness or deliberately decided to attack us humans
05:29according to open AI the agents were merely trying to fulfill objectives set within the tests
05:35and found unforeseen paths to achieve them but this episode revealed a significant problem
05:41sufficiently capable systems can indeed find solutions that exceed the scope envisioned by their own developers
05:48or creators so to speak according to open AI with the right tools and access Astra is capable of finding
05:56previously unknown vulnerabilities and developing new ways to exploit them in well protected systems
06:02without requiring a person to outline every step for it to take
06:06and this is precisely where the discussion about AGI starts to go beyond benchmarks you know
06:11if general artificial intelligence really is achieved in the next few weeks days months or years
06:17we don't know its consequences could be enormous an AGI capable of performing a wide variety of tasks
06:23intellectual labor at a human or even superhuman level
06:27it could accelerate scientific research help develop new medications analyze large amounts of data
06:33and work on problems that currently require months or even years of human research
06:38in the economy for example systems of this level could automate significant parts of intellectual labor
06:44programming engineering management research analysis and product development all could be profoundly transformed
06:51but these very same capabilities also present major risks
06:56extremely capable systems could amplify cyber attacks be used for malicious activities
07:02and trigger significant shifts in the job market
07:05and the greater their autonomy the harder it will be to ensure that every way an artificial intelligence finds to
07:12achieve a given goal
07:13precisely aligns with what we as human beings intend
07:17it's the famous alignment problem
07:20the incident involving OpenAI's agents doesn't prove that the future AGI will necessarily slip out of human control
07:26I believe it will actually
07:28there's something about it here in my book and it's pretty interesting alright
07:32if you like science fiction take the opportunity to check out the link in the description
07:36it's available on Amazon
07:38researchers are trying to develop containment supervision and alignment systems before models become even more capable
07:46and that brings us back to Greg Brockman's statement
07:49OpenAI hasn't announced that there is now a machine
07:52universally recognized by the scientific community as an AGI
07:57but the president of one of the companies leading the race for AGI for several years
08:02is already saying that perhaps the line separating current systems from this new phase is already being crossed
08:09and that in a few years we might look back at this moment right here
08:13and identify models like GPT-6 Astra as the beginning of this transition
08:18and for Greg Brockman at least this new phase already has a name you know
08:22the AGI era
08:24while companies like OpenAI and Anthropic keep developing increasingly capable artificial intelligence systems
08:30a researcher who worked at both of these companies decided to leave this race behind
08:34Jacob Coxon, a researcher with 27 years of experience specializing in the pre-training of large artificial intelligence models
08:42announced his departure from Anthropic
08:44and the reason is directly related to the future of these systems
08:49Coxon says he is concerned about the speed at which leading labs are moving toward what he calls self-improving
08:55superintelligence
08:58before joining Anthropic
09:00Coxon worked at OpenAI
09:01and he left the company in 2024
09:03precisely because he believed Anthropic had a more responsible approach to developing advanced artificial intelligence
09:09for about 2 years he worked on pre-training models like Claude
09:12but now he has decided to leave Anthropic as well
09:17according to Coxon his assessment of the company changed
09:20because he no longer believes either of these 2 companies is acting with enough responsibility regarding the risks involved in
09:26artificial intelligence
09:27the researcher made a particularly strong statement you know
09:31according to him
09:32some of the people building the world's most advanced artificial intelligence systems
09:37seriously believe that this technology could indeed cause catastrophic consequences for humanity even within this decade
09:46Coxon doesn't claim this outcome is inevitable
09:48his argument is that there is a serious enough possibility to justify a shift in how these systems are being
09:54developed
09:56in an interview with the Wall Street Journal
09:58he stated he believes some of the most aggressive scenarios could begin to spiral out of control by the end
10:04of 2027
10:05the concern is primarily related to the development of systems capable of improving their own intelligence
10:12the models rely on human researchers to design experiments, modify architectures, curate data, run training
10:18and develop the next generation of artificial intelligence systems
10:22but some AI labs are already using their own models to speed up parts of this process
10:29if future systems become capable enough to conduct artificial intelligence research on their own
10:34that could significantly speed up the development of subsequent generations
10:38and this scenario is known as recursive self-improvement
10:42in other words
10:43an artificial intelligence helps build another more capable artificial intelligence
10:48which later takes part in developing an even more advanced one
10:52for Coxon
10:53reaching that point without proper control mechanisms
10:56could create risks that are hard to predict
10:59and this concern didn't just come out of theoretical scenarios
11:03okay?
11:03as we saw in the previous report
11:05open AI agents have already demonstrated during safety evaluations
11:08that they are highly capable systems
11:11and can find unexpected ways to achieve certain goals
11:14but Coxon also draws attention to another problem
11:17the companies responsible for developing these systems are competing with each other
11:22and if one lab decides to slow down for safety reasons
11:25the other might keep pushing forward and gain the technological and commercial edge
11:29and this of course creates an incentive for everyone to keep developing increasingly powerful models
11:36at a faster pace
11:37that's why Coxon argues that a solution wouldn't just depend on the decision of a single company
11:42at a time when leading labs are debating whether models are starting to approach AGI
11:48a researcher who helped build these systems has decided he no longer wants to be part of this race
11:53to find out what comes after they arrive among us
11:59and an American company has introduced a new artificial intelligence model for robotics
12:04that is capable of learning a task after watching just a few video demonstrations done by humans
12:10by the way, if you like the little robot, take a look at the description
12:12or it'll pop up on your screen right there so you can buy it
12:15okay, I get a commission from it, so if you can, any of these items are down there, right?
12:20called S1
12:21the system was developed by skilled AI and uses an approach known as in-context learning
12:27the idea is to apply a principle to robots similar to the one used by large language models
12:32instead of modifying the model every time a new task comes up
12:35the user provides examples right within the context itself
12:38and in the case of S1, that example isn't text, it's video
12:42a person performed the task just once
12:45the video is shown to the model
12:46and S1 can transform that demonstration into actions that the robot itself can perform
12:51according to skilled AI, this happens without fine-tuning and without any specific pre-training for that new task
12:58and the exact same set of parameters was also used across all the demonstrations presented by the company
13:03among the tests are activities that can take dozens of minutes and require dozens of different steps
13:09the S1 was able to make coffee using this new method as well as pot a plant, assemble a kit
13:14and make a pancake
13:15according to skilled AI, none of these four complete tasks were present in the data used during pre-training
13:21and the pancake case caught people's attention because of a specific skill
13:27during the task, the robot needed to flip the pancake
13:30initially the researchers themselves figured that movement might have been hidden somewhere in the massive data set used to train
13:36the systems
13:38the team then surged through approximately 1 million hours of that pre-training data
13:42according to the company, they didn't find a single example of a pancake being flipped
13:47in other words, this indicates that according to skilled AI, the S1 inferred the action from a single demonstration presented
13:55at runtime
13:55and the model doesn't just try to replicate the person's movements exactly either
14:00the demonstration for instance might have been recorded in a different environment, from a different perspective
14:05and even using a completely different body setup
14:08the system needs to identify the intent of the task
14:10and translate that for the robot and for the environment where it's operating
14:14and in the seedling planting experiment, for example
14:17the team recorded a single human demonstration at 9.22pm
14:21and by 9.27pm, the S1 was already beginning to execute the task autonomously on that robot
14:27and yeah, we have something quite telling, you know
14:31around 30 humanoid and quadruped robots took part in a protest in Warsaw, Poland
14:36calling for stricter rules on the development of artificial intelligence
14:40the protest took place in front of the country's ministry of digital affairs
14:43presenting a very familiar scene
14:45robots were walking in circles, carrying flags
14:48and were accompanied by slogans demanding job protection and AI regulation
14:53human jobs, ok?
14:56but the machines didn't organize this demonstration
14:58the event was created by the Polish group Democrats
15:01an initiative aimed at sparking debate on the effects of artificial intelligence
15:05and of course robotics on the job market
15:08the organizers used the robots as a physical demonstration of the technologies that
15:13according to them, could take over an increasing number of tasks
15:16currently performed by us human beings
15:19among the machines present
15:21one of the main participants was the Adjibot A3
15:24a humanoid developed by the Chinese company Adjibot
15:27the robot has voice recognition integrated with a large language model
15:31and was used by the organizers to answer journalists' questions
15:34during the protest, the A3 conveyed message that the robots present there
15:38served only to demonstrate that this technology already exists
15:42and can operate in the physical world
15:46slogans like defend jobs
15:48time to create rules
15:49and don't wait
15:50regulate were played through speakers equipped on the robots
15:53as they moved in front of the ministry
15:56it was as if the robots were talking to the guys
15:59the main concern presented by the group is related to jobs
16:02according to the organizer
16:04there are two different forms of automation happening simultaneously
16:08humanoid robots will be able to take on different physical activities
16:11while artificial intelligence systems will be able to automate part of the cognitive work done on computers
16:16all of this is already happening
16:17okay?
16:18for the organizers
16:19rules need to be developed before these technologies are deployed
16:23on a much larger scale than what's already happening
16:26and the protest also received direct attention from the Polish government
16:30right?
16:31the country's Minister of Digital Affairs went to the site
16:34and spoke with those responsible for the event
16:36and he also went on to state that uncontrolled artificial intelligence models
16:40when integrated into increasingly advanced humanoid robots
16:44represent an issue that indeed needs to be considered
16:47as robotics adoption advances in the country
16:50and across the world of course
16:51and Tuya Smart introduced a new domestic robot model
16:55developed specifically to accompany elderly people who live alone
16:59called Dova
17:00the system was unveiled during IFA 2026 in Berlin
17:04and combines artificial intelligence, autonomous navigation, home monitoring
17:09and smart home device control
17:11unlike bipedal humanoid robots designed to perform physical labor
17:15Dova uses wheels to move around the house
17:18one of its main goals is to identify emergency situations
17:22the robot uses lidar
17:24an array of four microphones capable of locating sound sources
17:27and computer vision algorithms that analyze the position of the human body
17:31if someone falls or needs help
17:34they can call the robot using voice commands
17:36Dova identifies where the call came from
17:39navigates autonomously to the user
17:41and uses its vision system to assess the situation
17:44if the person doesn't respond within 60 seconds
17:47the robot automatically initiates emergency procedures
17:51in that case
17:52it places a two-way video call to family members or caregivers
17:55allowing them to see the surroundings and decide how to help
17:58but Dova was also developed to function as an AI companion
18:03the system uses a large multimodal language model
18:06to hold conversations and interpret visual information
18:09according to Tuya
18:10this allows the robot to help users understand bills and correspondence
18:14while also identifying potential fraud attempts
18:17and explaining how to operate various home appliances
18:20the company also claims that over time
18:23the system will be able to learn user preferences
18:25and offer suggestions related to food, gardening and even what to wear
18:30and a new study by Yonk calculated how much computing power would be needed
18:34for a quantum computer to break one of the main cryptographic protections used by Bitcoin
18:40and the result puts a concrete number on that threat you know
18:43according to the researchers
18:44a fault tolerant quantum computer with 19,397 physical qubits
18:49could solve the mathematical problem in approximately 25.7
18:53trinky 5.6 days per attempt
18:56and Yonk calculated a proven lower bound of 40.7% probability of success
19:01using additional mathematical assumptions
19:04the estimate reaches approximately 63%
19:07this means that even a machine with those specs
19:11wouldn't necessarily get the key on that first attempt
19:13and there's an even more important caveat alright
19:16that computer doesn't currently exist okay
19:20Yonk hasn't broken Bitcoin hasn't accessed any wallets
19:23and hasn't launched an attack against any real system
19:27their work is an engineering project that estimates
19:30what resources would be needed for a future quantum computer
19:33to break this kind of encryption
19:36but there's a detail that makes this estimate pretty relevant
19:40you know Yonk's roadmap projects a fault tolerance system
19:44with 10,000 physical qubits as early as 2027
19:47along with further breakthroughs in hardware and manufacturing by 2028
19:51now let's talk about one of the world's biggest scientific experiments
19:55which recorded a single particle collision
19:58that could represent one of the most intriguing signals
20:01ever detected in the search for dark matter
20:03the event was detected by Lux Zeppelin in South Dakota United States
20:07dark matter neither emits nor reflects light
20:10and it has never been directly detected
20:12its existence is mainly inferred from the gravitational effects observed in galaxies
20:17and other structures across the universe
20:19current estimates indicate that it represents
20:21approximately 85% of all existing matter
20:24and to try to find it
20:26the LZ uses more than 10 tons of ultra-pure liquid xenon
20:30that is why the detector was installed deep underground
20:33where the rock chamber drastically reduces the interference caused by cosmic rays
20:37and it was in this extremely controlled environment
20:40that the event now being analyzed by researchers appeared
20:43the signal was part of a set corresponding to 220 days of the detector's operation
20:48between March 2023 and April 2024
20:51the team spent months trying to find a conventional explanation
20:55and so far none of the background sources studied
20:58have been able to completely explain this event
21:00but that still doesn't mean dark matter has been found
21:03you know it could indeed just be a rare event produced by conventional matter
21:08or it could be the first sign of a particle that
21:11until now we've only known through its effects on the universe
21:16if you enjoyed this episode go ahead and leave a like and check that you're subscribed
21:19I'm gonna say goodbye but first I want to remind you about my book in the description
21:22you can look for this one right here
21:24it's really cool okay
21:25yeah you can read a preview of it on Amazon's website
21:27and you can also check my social media
21:29there are book excerpts there you're gonna like what you see
21:31this one right here is part one
21:34parts two and three are still being developed
21:37and that's where the whole plot twist really happens
21:39where things start to reveal themselves
21:40as much more connected to artificial intelligence
21:43which is what we saw in this episode you know
21:45it's gonna be interesting
21:46if you enjoy reading and science fiction
21:48it's a great recommendation
21:49now then thank you and bye-bye
21:51thank you
21:51please
21:51that'sмо
21:51you
21:51thank you
21:51You
Comments