terminalcommand

I explicitly asked LLM to write abstract away with a DSL and it did that. LLMs are good at this too.

LLMs are able to write racket. I have a fairly large project I've been vibecoding for the last 6 months. Models predating deepseek flash 0731 used to make paren mistakes and had a hard time. But now Opus 5.5 for example does not make any mistake writing clear racket code.

Corporate lawyer at big law firm here. I believe the model will be like this: we will review your 1000 agreements but with AI. AI is generally accurate but has some risks, if you agree we will charge you less. The client will agree. Instead of spending 300-400 hours to review the ageeements AI will do instantly. A lawyer will spend the time on reviewing the red flags identified and draft a report. We already do the same with translation. Before computer translation became good (i.e pre-Deepl era), we used to translate documents (cease and desist letters, court petitions etc.) ourselves. Now we offer AI translation free of charge. If client wants reliance on the translation we charge a small bit. Lawfirms are also under pressure for meeting internal budgets. Most clients ask for capped fees. If you charge 10k for a job to a client and the hourly costs ramp up to 20k, the management criticises the partner, the associates become unwilling to work for the partner (as their hours get written off). So with AI law firms can bet more competitively as the cost (human attorney hours) are reduced.

Lawyer here. I used to trust Claude as hallucinations are near non-existent now. However for large volume tasks such as due diligence exercises, they still happen. We also tried Legora's tabular review, there were also numerous halucinated provisions in our due diligence exercise.

To ration insulin you limit eating carbs. If you don't eat, you need less insulin.

Treatment for DKA as far as I kmow is to give glucose insulin and potassium.

For example if you have a stomach bug and can't keep down anything you are at risk of developing DKA even if your BG is normal.

The thing is how do we know we are starving the body of fuel. Most diabetics (me included) are told not to exercise if bg is higher than 250mg/dl, but that is it. We hit over 250 mg/dl frequently and sometimes stay there for multiple hours (infusion set problems, fatty food etc.).

T1D athletes report that they eat a lot and take a lot insulin before/after training and they say they get tired very easily if they don't. They thought this as restoring the glucogen reserves in muscles but maybe it is about fueling the body.

I sometimes feel uneasy even my bg is okay and it had led me to panic attacks. Now when I think it could be raised ketones, or just anxiety.

Knowing the ketones would be helpful.

https://journals.sagepub.com/doi/10.1177/11795514261431390

There is a huge gaming community who use these processors to emulate gaming consoles and windows games. I am still amazed I can emulate PS2 and Switch games with upscaled resolution on my S25 Ultra.

Ketoacidosis can also happen with stable blood sugars if you don't get enough insulin. There were people rationing insulin due to cost and some of them died of DKA even if their blood glucose was perfectly stable. Another possibility of DKA with good blood sugars is when you are sick. I think this data would be helpful.

Yes, this is true. There were also news about some Americans rationing insulin due to cost. Their blood sugar levels were normal, but they had high ketones and unfortunately died of DKA. When I got first diagnosed, we were given ketone urine strips and told to use them, with the advance of glucometers, we forgot about ketone testing.

Minimed 780g user here, it is not perfect. For a true closed loop we need a system that can give glucagon as well. The current closed loop only gives you insulin to lower your blood glucose. When it gives too much it wakes you up and asks you to treat it. You also have to accurately count carbs of meals you eat, enter when you exercise. The sensors are also not quite reliable so sometimes it freaks out and you calibrate and go manual.

For a closed loop we need insulin and glucagon in conjunction to keep the bg stable without user intervention.

AFAIK, apple does not release drivers open source, asahi is a reverse-engineering endeavour and does not support GPU. For nvidia, there are both proprietary and open-source linux drivers. CUDA and inference works on linux with nvidia. I would recommend checking out this video of Alex Ziskind to shop for a computer to run local LLMs: https://www.youtube.com/watch?v=mevUEQcumzU&t=224s. TL;DR besides Apple he recommends, DGX Spark, Tenstorrent Wormhole N300, AMD Radeon 7900 and NVIDIA RTX 5090.

I built open baffle speakers based on measurements and discussion I had with Claude. I think it is really good.

I am a novice, maybe that's why I liked it.

I think you are confusing Erdogan with Imamoglu. Imamoglu has photographs, classmates, there is no doubt that he attended college. His diploma was anulled because he transferred initially from a different college. He spent 4 years at the Istanbul uni, attended classes, passed the exams, there is no doubt in that.

On the other hand, Erdogan does not have a single photograph during his university years, no classmates to back his story. He started a two year degree, but there is no evidence he attended a four year program. A public notary issued a same as original certification on a disputed document. The original diploma of erdogan cannot be found. Looking at the date of the diploma, the university faculty didn't even exist yet.

Another T1D here. I do not have a compatible pump with looping. But I'd like to dip my toe into openaps.

I use a cgm (libre2).

Can I use autotune to tune my carb ratio, basals etc. without looping? How was your experience in this?

Do I have to use nightscout to run autotune?

Have you tried Claude.ai. In my experience on computer science topics, the LLMs are very good. Because they have been trained on a vast amount of information online. I just had a nice conversation about mutexes and semaphores with claude and was able to finally grasp what they were.

I do not know if this is the case for example for mathematics or sciences.

No, this is not accurate in my trials. I use Claude.ai daily. If you ask questions on niche topics or dive down too deep, it says that resources on the topic are limited and you should consult a book.

Will this work on older cards such as RX 570? Does anyone know?

Isn't microsoft phi specifically trained for Python? I recall that Phi 1 was advertised as a Python coding helper.

It's a small model trained only by quality sources (ie textbooks).

This was a very cool read. Programmers were programmers even back in the day of Mark I.

It is cool to see that they dabbled in natural language processing back then. This is years before Eliza and they were working on generating English prose based on English grammar. Very impressive!

The music generation program they wrote is equally impressive. The recording that was playing shows that they were adept enough to time events in the computer so good that they could playback songs. This was back in the early 1950s.

There are ways around this. For example set a property tax from second home on. Do not tax the primary residence. Or set income brackets. If poorer people live in their own homes they don't pay property tax.

You do not have to care if it is occupied. Tax the property tax. If the owner does not rent out the property, they'll pay out of pocket.

That is the very problem we are facing in Turkey :). The municipality determines the value of housing in a neighborhood each year. That is taken as a basis for property taxes and transaction taxes. The municipality assessed value is somewhere near 1/20th of the value of an average flat. So, almost no tax gets collected :(.

In countries with high inflation purchasing real-estate and keeping it vacant is an inflation hedge. Plus, you also benefit from low interest rates and get free money if your government allows it.

I live in Turkey. We had 80% p.a. inflation, where the government decided to lower the interest rates even further. Our president said Interest rates are the cause of inflation and if we lowered interest rates inflation would go down. State banks gave out house loans with 12% p.a. interest where the inflation rate was above 80% p.a.

A lot of Turkish people got their free money from the bank and invested in real estate. In Turkey, everyone evades tax and property taxes are not really collected. This in turn fueled inflation even more, sky-rocketed inequality and caused the worst housing crisis.

That is why I am convinced that property taxes are a must.

The solution is actually simpler, set a property tax that would hurt if the buildings became vacant. For example if you pay 1% of the buildings value as property tax each year, it would make enough incentive to rent it out or sell if you don't need it. The proceeds can be used for building public housing projects or helping the homeless. Property tax was invented for this very reason.

Well, I try to be optimistic and work with the models.

It's like when we first learned to code. Did syntax errors scare us, did nullpointer exceptions, runtime panics scare us? No, we learned to write code nevertheless.

I use LLMs daily to enhance my productivity, I try to understand them.

Providing context and assigning roles was a tactic I was taught in a prompt writing seminar. It may be a totally wrong view to approach it but it works for me.

With each iteration the LLMs get smarter.

Let me propose another example. Think of the early days of computing. If you were an old school engineer who only relied on calculations with your trusted slide rule, you would critise computers because they made errors, they crashed. Computing hardware was not stable back then and the UI were barely usable. Calculations had to be double checked.

Was investing in learning computing a bad investment then? Likewise investing in using LLMs is not a bad investment now.

They won't replace us, take our jobs. Let's embrace LLMs and try to be constructive. We are the technically inclined after all. Speaking of faults and doom is easy, let's be constructive.

I may be too dumb to use LLMs properly, but I advocate for AI because I believe it is the revolutionary next step in computing tools.

Well I told you I was bad at math. I wrote a faulty prompt. But chatgpt understood my instructions perfectly. That was the argument I was trying to demonstrate.

The problem lied between the chair and the computer.

We have to learn how to use LLMs.

Understood. I tried your prompt again and it seems it understood it.

Here is my experiment: https://chat.openai.com/share/98cae2bf-a7a6-42e7-b536-f3671c...

I gave minimum context like this: "I have a history exam. You are an expert in British royal history. List me the names of 20 kings and queens in England."

The answer was: "Certainly! Here's a list of 20 kings and queens of England:

1. William the Conqueror 2. William II (Rufus) 3. Henry I 4. Stephen 5. Henry II 6. Richard I (the Lionheart) 7. John 8. Henry III 9. Edward I (Longshanks) 10. Edward II 11. Edward III 12. Richard II 13. Henry IV 14. Henry V 15. Henry VI 16. Edward IV 17. Edward V 18. Richard III 19. Henry VII 20. Henry VIII"

I disagree. AI in 1960s relied on expert systems where each fact and rule was handcoded by humans. As far as I know LLMs learn on their own on vast bodies of text. There is some level of supervision, but it is bot 1960s AI. That is the reason we get hallucinations as well.

Expert systems are more accurate as they rely on first order logic.

This reminds me of movies shot in early times of the internet. We were warned that information on the internet could be inaccurate or falsified.

We found solutions to minimize wrong information for example we built and maintain Wikipedia.

LLMs will also come to a point where we can work with them comfortably. Maybe we will ask a council of various LLMs before taking an answer for granted, just like we would surf a couple of websites.

Thanks for trying. With the prompt I provided chatgpt was able to play and understand the win condition. However the moves were stupid.

If I changed the prompt and removed the word win, it did not understand the win conditions as well.

Here were my experiments: https://chat.openai.com/share/f02fbe93-dfc5-4d8a-9cf3-b1ae34...

I even exclaimed you are lousy at Tic Tac Toe to GPT.

It seems that GPT3.5 struggles to play visual games.

It is marvelous that a statistical word guessing model can get so far though :).

Well it is a bit like satire. You have to explain the universe for an unspecialized GPT, like you would do to a layman. There are custom gpts that come preloaded with that universe explanation.

In addition, do not ask facts to an LLM. Give a list of let's say 1000 kings of a country and then ask give 20 of those.

If you ask 25 kings of some country, you are testing knowledge not intelligence.

I see LLMs like a speaking rubber duckie. The point where I write a successful point is also the point where I understand the problem.