Number two, context window and knowledge. Both 01 and 01 mini offer a 128,000 token context window and an October 2023 knowledge cutoff, but it’s worth noting that some competitors like Claude for Enterprise already offer even larger context windows of 500,000 tokens. However, for most applications, 128,000 tokens is more than sufficient, and with the October 2023 knowledge cutoff, 01 is able to work with relatively new information from the get-go. Number three, test accuracy. In standardized tests, 01 consistently outperforms PhD-level accuracy across physics, biology, and chemistry. And when compared to GPT-4-0, 01 outperforms by a long shot, proving it to be a valuable upgrade for researchers, educators, and students in these fields.

Altogether, the model excels in areas like scientific research and education, where complex problem-solving scenarios arise across various scientific disciplines. Number four, competitive programming. For software developers and computer science enthusiasts, 01’s coding abilities are particularly noteworthy, with the model ranking in the 89th percentile on competitive programming questions. This makes 01 capable of tackling complex algorithmic challenges and optimizing code, with software development teams possibly leveraging 01’s capabilities to streamline, solve, and develop more efficiently, but with the trade-off of response time, as 01 has much longer wait times for model outputs. In testing, 01 scored significantly higher than GPT-4-0, being about four times harder to jailbreak, demonstrating a stronger adherence to safety protocols than both GPT-4-0 and Claude 3.5 Sonnet.

For organizations dealing with sensitive information such as healthcare, finance, and government, 01 offers more data security and a decreased risk of malicious outputs or brand damage. Number six, speed. The 01 family consists of two main models, with the first being the 01 flagship and the second model being the 01 mini, which is smaller, faster, and cheaper. In fact, because both 01 and 01 mini use around 10 times more compute than their respective GPT-4-0 predecessors, the 01 flagship model is about 30 times slower than GPT-4-0, while 01 mini is about 16 times slower than GPT-4-0 mini.

All in all, this makes 01 essentially unfit for most production applications as of now, but how much does it cost? For now, OpenAI’s 01 comes at a premium of $15 per million input tokens and $60 per million output tokens, while the more budget-friendly 01 mini is priced at $3 per million input tokens and $12 per million output tokens. In comparison, GPT-4-0 costs just $2.50 and $10 for input and output tokens, respectively, while GPT-4-0 mini is significantly cheaper still at just $0.15 and $0.60 for input and output tokens. Finally, Claude 3.5 Sonnet falls in between, being priced at $3 per million tokens of input and $15 per million output tokens.

And regarding availability, 01 models are currently accessible through ChatGPT Plus to enterprise users, plus via API for developers on tier 5 of API usage. However, there are usage limitations in place for a limit of only 30 messages per week when using the 01 preview and 50 messages per week for the 01 mini. After reaching these limits, users are required to switch to GPT-4-0 models, with these limitations potentially impacting heavy users, thus requiring strategically allocated usage or alternative solutions for high-volume use cases. Overall, 01 seems to fit niche use cases for STEM professionals needing advanced reasoning and complex coding with a prioritization on safety, but not requirements on speed.

Meanwhile, a new robot named Alex has just been unveiled. Powered by advanced artificial intelligence, this state-of-the-art humanoid robot can seamlessly integrate into a wide range of industries. Plus, Alex’s ability to tackle repetitive tasks and complex problem-solving makes it both a scaling and productivity solution in business. For instance, in manufacturing, Alex can serve with precision and speed using its 19 degrees of freedom and high-speed joints that are capable of moving at 9 radians per second. And Alex can handle a diverse array of tasks, having the ability to work continuously while carrying a payload of up to 10 kilograms.

But Alex doesn’t stop at the assembly line. In the world of logistics, it brings a human-like touch to sorting packages and unloading irregular cargo without requiring actual human intervention. With its 300-degree range of motion wrists, Alex can easily manipulate and move items, even in challenging or awkward positions. Plus, its backdriving torque ensures the robot operates safely, with precise control over its use of force to reduce the risk of damage to delicate items. And with its agnostic end-effector and lower body, Alex is adaptable, offering flexibility for use cases across various environments and industries.

But another humanoid is coming to the workplace soon, as Neura just released new footage, showing its 4NE1 humanoid robot that’s designed to integrate into daily life at home and at work. Running on Neura’s cognitive AI platform, this humanoid aims to enhance human-machine interaction with more intuitive and natural communication, using its hardware to see, hear, and sense touch. And one of the ways it does this is with advanced 3D vision, allowing the 4NE1 to recognize objects, environments, and gestures to respond to its surroundings in a human-like manner. To enable this, the robot’s force-talk sensors provide its sense of touch, which is crucial for performing dexterous real-world tasks.

And because safety is paramount, the 4NE1 features a touchless human detection sensor, ensuring it operates safely around people too. Standing at 180 centimeters tall and weighing 80 kilograms, 4NE1 easily fits into spaces originally designed for humans. Plus, it can carry a payload of up to 15 kilograms and can move at a speed of 3 kilometers per hour, making it capable, albeit a bit slow. But where it lacks in speed, it makes up in dexterity with its interchangeable forearms that further enhance its adaptability, allowing it to switch between multiple functions effortlessly.

Furthermore, because the robot is built on top of the Neuraverse platform, 4NE1 is positioned to grow in the humanoid robot market, which is expected to accelerate by double-digit percentages. And Neura’s use of Nvidia’s Omniverse for training ensures it can remain at the cutting edge as new AI paradigms spring up, allowing 4NE1 to automate an increasing number of repetitive tasks, requiring low dexterity over the next three to five years. Altogether, robots like Alex and Neura’s 4NE1 are likely coming to automate a large percentage of unskilled jobs, taking care of repetitive and dangerous tasks so that humans can focus on more meaningful ones.

[tr:trw].