CodeInvasion
This is truly a terrible accident. Given the flight tracking data and the cold, winter weather at the time, structural icing is likely to have caused the crash.
Ice will increase an aircraft’s stall speed, and especially when an aircraft is flown with autopilot on in icing conditions, the autopilot pitch trim can end up being set to the limits of the aircraft without the pilots ever knowing.
Eventually the icing situation becomes so severe that the stall speed of the ice-laden wing and elevator exceeds the current cruising speed and results in a aerodynamic stall, which if not immediately corrected with the right control inputs will develop into a spin.
The spin shown in several videos is a terrifying flat spin. Flat spins develop from normal spins after just a few rotations. It’s very sad and unfortunate that we can hear that both engines are giving power while the plane is in a flat spin towards the ground. The first thing to do when a spin is encountered is to eliminate all sources of power as this will aggravate a spin into a flat spin.
Once a flat spin is encountered, recovery from that condition is not guaranteed, especially in multi-engine aircraft where the outboard engines create a lot of rotational inertia.
Valve is a unique company with no traditional hierarchy. In business school, I read a very interesting Harvard Business Review article on the subject. Unfortunately it’s locked behind a paywall, but this is Google AI’s summary of the article which I confirm to be true from what I remember:
According to a Harvard Business Review article from 2013, Valve, the gaming company that created Half Life and Portal, has a unique organizational structure that includes a flat management system called “Flatland”. This structure eliminates traditional hierarchies and bosses, allowing employees to choose their own projects and have autonomy. Other features of Valve’s structure include:
- Self-allocated time: Employees have complete control over how they allocate their time
- No managers: There is no managerial oversight
- Fluid structure: Desks have wheels so employees can easily move between teams, or “cabals”
- Peer-based performance reviews: Employees evaluate each other’s performance and stack rank them
- Hiring: Valve has a unique hiring process that supports recruiting people with a variety of skills
It took Hawking minutes to create some responses. Without the use of his hand due to his disease, he relied on the twitch of a few facial muscles to select from a list of available words.
As funny as it is, that interview, or any interview with Hawkins contains pre-drafted responses from Hawking and follows a script.
But the small facial movements showing his emotion still showed Hawking had fun doing it.
I am an LLM researcher at MIT, and hopefully this will help.
As others have answered, LLMs have only learned the ability to autocomplete given some input, known as the prompt. Functionally, the model is strictly predicting the probability of the next word+, called tokens, with some randomness injected so the output isn’t exactly the same for any given prompt.
The probability of the next word comes from what was in the model’s training data, in combination with a very complex mathematical method to compute the impact of all previous words with every other previous word and with the new predicted word, called self-attention, but you can think of this like a computed relatedness factor.
This relatedness factor is very computationally expensive and grows exponentially, so models are limited by how many previous words can be used to compute relatedness. This limitation is called the Context Window. The recent breakthroughs in LLMs come from the use of very large context windows to learn the relationships of as many words as possible.
This process of predicting the next word is repeated iteratively until a special stop token is generated, which tells the model go stop generating more words. So literally, the models builds entire responses one word at a time from left to right.
Because all future words are predicated on the previously stated words in either the prompt or subsequent generated words, it becomes impossible to apply even the most basic logical concepts, unless all the components required are present in the prompt or have somehow serendipitously been stated by the model in its generated response.
This is also why LLMs tend to work better when you ask them to work out all the steps of a problem instead of jumping to a conclusion, and why the best models tend to rely on extremely verbose answers to give you the simple piece of information you were looking for.
From this fundamental understanding, hopefully you can now reason the LLM limitations in factual understanding as well. For instance, if a given fact was never mentioned in the training data, or an answer simply doesn’t exist, the model will make it up, inferring the next most likely word to create a plausible sounding statement. Essentially, the model has been faking language understanding so much, that even when the model has no factual basis for an answer, it can easily trick a unwitting human into believing the answer to be correct.
—-
+more specifically these words are tokens which usually contain some smaller part of a word. For instance, understand and able would be represented as two tokens that when put together would become the word understandable.
I am a pilot and this is NOT how autopilot works.
There is some autoland capabilities in the larger commercial airliners, but autopilot can be as simple as a wing-leveler.
The waypoints must be programmed by the pilot in the GPS. Altitude is entirely controlled by the pilot, not the plane, except when on a programming instrument approach, and only when it captures the glideslope (so you need to be in the correct general area in 3d space for it to work).
An autopilot is actually a major hazard to the untrained pilot and has killed many, many untrained pilots as a result.
Whereas when I get in my Tesla, I use voice commands to say where I want to go and now-a-days, I don’t have to make interventions. Even when it was first released 6 years ago, it still did more than most aircraft autopilots.