I’m pretty sure this is because they don’t want Gemini saying things like, “based on my stored context from our previous chat, you said you were highly proficient in Alembic.”
It’s hard to get a principled autocomplete system like these to behave consistently. Take a look at Claude’s latest memory-system prompt for how it handles user memory.
It could be that the instruction was vague enough ("never mention user_context unless the user brings it up", eg) and since the user never mentioned "context", the model treated it as not having been, technically speaking, mentioned.
I agree, this might just be an interface design decision.
Maybe telling it not to talk about internal data structures was the easiest way to give it a generic "human" nature, and also to avoid users explicitly asking about internal details.
It's also possible that this is a simple way to introduce "tact": imagine asking something with others present and having it respond "well you have a history of suicidal thoughts and are considering breaking up with your partner...". In general, when you don't know who is listening, don't bring up previous conversations.
The tact aspect seems like a real possibility. In a world where users are likely to cut&paste responses it can't really be sprinkling in references like this.
There's typically a button to toggle it on/off somewhere near the steering wheel.
I got a new car a couple of months ago that has this, and was experiencing similar issues on rural Texas farm roads. I just had to look up where mine was and turn it off temporarily.
I have an Audi - the button to turn it off and on is at the tip of the turn signal stalk. You can toggle it with your pinky without removing your hands from the steering wheel or eyes from the road.
I also have a Subaru - turning it off or on requires three levels of menu navigation, and you have to use the center console touch screen.
I suppose they do their testing and validation in drastically different environments!
After recent events centered on the Media Lab, I know there is a tremendous sense of pain, sadness and disappointment across campus and throughout our global community. We all want answers.
For now, I write to share two important updates.
Last Saturday, we retained the law firm Goodwin Procter to perform a thorough investigation of the facts surrounding Jeffrey Epstein’s interactions with MIT. This work has already begun. We have asked the firm to explore all donations received by MIT, both those that came directly from Jeffrey Epstein and his associated foundations, and any donations that may have been received at the direction of Jeffrey Epstein. In addition, the investigation will cover who at MIT may have been aware of the donations. We have instructed Goodwin Procter to follow the evidence where it leads, and we are counting on this independent investigation to ascertain the facts. As I noted on Saturday, the firm will report back to me and to the Executive Committee of the MIT Corporation, MIT’s governing board.
If you have information relevant to the investigation that you would like to share with the law firm, please email REDACTED.
To be thorough, such an investigation will take time, on the order of a month. Although media reports will continue to focus on the issues during this time, I am hopeful that the MIT community will avoid forming a final judgment before the process is complete and will respect the privacy of members of our community who may have become involved in this matter in the course of doing their jobs for MIT.
Once we have the results, and once our separate internal review of our current processes on gift acceptance is complete, we will be able to understand what happened and what needs to change.
Supporting the members of the Media Lab must now be at the center of our attention. The Provost has asked the Dean of the School of Architecture and Planning to work with the Media Lab community on interim leadership for the lab and a search for the next director. The interim plan will be finalized shortly, and we will share the news as soon as possible, along with other plans to support the lab community.
I know this is a difficult and disorienting moment. But I have profound faith in the MIT community to learn from these events and find a constructive way forward.
The main rumor these days (well, it's been a rumor for quite a few years) is that they'll be dropping Intel and switching to house-designed ARM chips across the board.
Well that would be the case if they didn't introduce Mac Pro. And now it is obvious that Apple don't want to do that, at least there is no way Apple is going to design a 250W CPU just for the Mac Pro.
Excessive airloads on the stabilizer may require effort by both pilots to correct the
mis-trim. In extreme cases it may be necessary to aerodynamically relieve the
airloads to allow manual trimming. Accelerate or decelerate towards the in-trim
speed while attempting to trim manually.
To be clear, that's a link that agrees with me that it can be impossible, not a link that says they just needed to work harder.
It's also from a 20 year-old manual, and isn't present on newer manuals, and pilots haven't trained on it in decades. Not appropriate to expect pilots to know, and possibly not helpful here: the pilots were low altitude, so allowing the nose to drop even further (to relieve aerodynamic load) may have been deadly too.
It’s hard to get a principled autocomplete system like these to behave consistently. Take a look at Claude’s latest memory-system prompt for how it handles user memory.
https://x.com/kumabwari/status/1986588697245196348