mostly AI / ML / Quant ramblings from a multiple startup founder CTO guy

Helsinki, Finland
Joined March 2008
What the hell, never going to buy a Samsung tv
did you know your samsung tv has a service called rmdemon (that's not a typo, demon not daemon) that starts a PACKET LOGGER on your network. what possible legitimate use case could there be here, samsung?
2
265
mikkom retweeted
😁
Kimi K3.1 Pro vs GPT-6 Astra!
1
1
32
5,886
Finally new ATH, this month has been very good for my strategy.
2
12
495
Trump: "Russia has been fine on Ukraine"
“I saw hacksaws, chainsaws, axes, and machetes there. Everything was covered in blood.” Ukrainian military medic Yuriy Armash describes the terror inside Russian torture chambers. Ukrainian military medic Yuriy Armash spent more than a year in Russian captivity. Among other places, he was held in one of the most notorious torture facilities in the Kherson region, established by Russian occupiers inside a seized police station in Nova Kakhovka. There, he learned the stories of many people and witnessed how the Russians’ system of terror against local civilians operated. How Russians torture prisoners “The interrogations began on the very first day. They were conducted by the same FSB officers. During interrogations, they covered my face and handcuffed my hands behind my back. They asked where the weapons were, how many people were in our reconnaissance group, where our positions were, where we had come from, and what our mission was. They used an awl. They would pull the skin on my neck, pierce it, and leave the awl hanging there so I could feel it. Because I wasn’t telling them what they wanted to hear, since I couldn’t tell them something that didn’t exist, they started using a knife. They stabbed me in the leg. They would take a knife and press it into my upper leg until they felt the blade reach the bone. When the Russians realized they weren’t going to get an answer, they used a ‘Tapik,’ a military field telephone. They attached its wires to my ears and kept turning the crank, sending electric shocks through my body until I started screaming.” How Russian occupiers abused civilians “Starting April 20, they began taking me out to provide medical assistance to civilians who had been severely beaten by the Russians. I remember treating a woman who was about 40 years old in one of the two punishment cells in the basement. I bandaged her head because she had been beaten with a rubber baton. There were all kinds of injuries. They broke people’s legs. Men under the age of 35–40 were treated particularly brutally.” A room filled with blood-covered torture instruments “I never expected the level of brutality I witnessed after February 24. Although I had seen many things before that, I was honestly shocked. At the Nova Kakhovka police station, I entered rooms where I saw things that absolutely terrified me. I saw hacksaws, chainsaws, axes, and machetes. Everything was covered in blood. The room with the blood-covered torture instruments was on the second floor of the police station, in the left wing. It was just an ordinary office. But there were also more secret rooms in the building where they never even took me. At the police station, the occupiers also forcibly removed tattoos they didn’t like from the bodies of detained civilians. Some had their tattoos cut out with knives. Others had them burned off with potassium permanganate. Russian occupiers raped even children “When Buryats and Kalmyks were stationed at the police department, they frequently entered the cells of young women under the age of 25–30. They even sorted the women by age and placed them in separate cells. From time to time, women were taken out of their cells and brought to another wing. The young women were taken there at night and raped. When the Russians’ rotation was coming to an end, they sometimes entered the cells and raped the women right there. My cell was nearby. I could hear everything clearly… The worst case involved a 14-year-old girl whom the Russians were holding captive. She tried very hard to hide the fact that she had been raped. One day, when we were taken outside for a walk, I noticed that she was bleeding. I asked her what had happened. She didn’t want to answer. I asked, ‘Did Russian soldiers do this to you?’ She silently lowered her head. I assured her that no one except me would ever know about it. After that, she answered: ‘Yes.’”
2
247
After long time with using codex as the daily driver I'm now back on plus account. After suddenly getting quotas cut to half, reading all the comments about how Astra is slower and/or dumber again today and how sol is eating quota like a pacman I'm quite happy that I don't have to care anymore.
1
3
395
First full year of my current production system done. Not spectacular results but quite okay with all the technical problems I experienced. This is the problem with totally custom architecture. CAGR: +63.48% Sharpe: 1.68 Sortino: 2.58 MAR: 2.24 Max DD: 28.35% (BAD - due to a bug) Pictures attached are: The actual 1y IBKR screenshot and current status with some extra metrics. (The system is mostly daily level long/short stock portfolio strategy that also does some intraday trades for risk management purposes) I had some technical problems (it's a custom platform) and system bugs along the way that caused that horrible drawdown that both ate my early year profits to that point and that also drives the yearly metrics down considerably. Fortunately when the bug was fixed the system was able to dig itself out of the hole and accomplished a new ATH. I did many tweaks during the year, introduced many new elements, tuned risk management, changed portfolio size boundaries and did many other improvements. Basically it has been continuous research loop. However I haven't made any changes to the system after summer. I just don't seem to be able to find any improvements that would benefit long term reward/risk ratio so the system. I already did a separate post about the problem period / bug I had but basically I introduced a new element to the system (which I still have btw). I basically had a very subtle bug that caused continuous daily losses. I kept the system running as I thought it was a normal bad period but at some point I just had to pull the plug, had a very thorough inspection and found a subtle nasty bug regarding data fetch/build timing and fixed it. The system continued it's normal work and results can be seen from the curve. The drawdown at the end of the curve (currently still about 5% into that) is just a normal system fluctuation. The architecture of the system is almost the same to what it was year ago but there have been a many small changes and fixes to problems that you only find when you are full live on decent account. Happy to continue from here, I expect next full year to be much better as the system, is far better tuned now and most of the technical problems are hopefully fixed.
11
6
1
148
12,659
mikkom retweeted
Can you imagine if the U.S., in the middle of the fight against the Nazis, had switched sides and allied ourselves with the Germans. That's exactly what's happened with Russia. We have not only betrayed Ukraine and Europe, but we've betrayed all who have fought and died for freedom.
From Russia, with Love…
564
2,914
58
19,901
278,025
Just noticed that I still have about a week of old x20 codex account available and 2 resets... After that it's plus account and claude code migration for me. Time to put codex to full allocation doing some research until the quotas run out.
2
4
755
So sad but true
511
Another angle to this is that dogs effectively stumbled onto what is for them the equivalent of artificial superintelligence 20,000 years ago. The Singularity already happened for dogs. They live in a world they cannot understand, surrounded by miraculous wonders they take entirely for granted, their every need seen to by entities whose thoughts they cannot begin to fathom. The result is that there are about a billion dogs on the planet now. Most of them are pretty happy. Almost none of them need to work. It worked out pretty well for them.
555
2,251
437
22,215
1,578,929
mikkom retweeted
“Quantum elliptic curves might break crypto’s public key cryptography.” “No they won’t, do you know anything about cryptography?” “No, do you?” “No.”
56
256
29
3,337
80,767
Whoa. Super valuable resource if this is what it says it is.
If you're out there building agentic agents and algos you know how expensive data is. 23 years of 1 minute data is free from @UArkansas. API is rate limited to 300rpm. @MrMilkTrading hfdatalibrary.com
1
3
1,008
hmmm seems to be IEX only, not SIP data and prior IEX pi Trading data which I haven't tested myself but multiple reviews say it's inaccurate. So most likely not the most accurate data source. Would keep away for serious research.
3
389
Not sure who in OpenAI planned these EU prices: 1x plus: 23 5x pro 100: 103 eur (20.6 scaled to plus) 10x pro 200: 229 eur (22.9 scaled to plus) So currently "100" is the cheapest subscription per token, not 200.
4
1
457
so basically you get no benefit from higher tiers. it just scales directly based on price. therefore it's always best to get the cheapest plan and only upgrade if you hit the limits and immediately cancel the sub after you have paid for the period. especially as openai only bills you full price for the period you pay for so if you can cope with smaller accounts for some time you get month cheaper. Thanks for the explanation
I’ll explain the new Pro 200 plan differently, before I start live tweeting from DevDay on things that are going out! Today we are going to ship a number of things that increase what you can do across the Plus and Pro plans. A lot of compute is online for this increase. As we increase the floor, we are changing the relative difference between plans to be Plus = 1X Pro 100 = 5X Pro 200 = 10X and we are reopening subscriptions for Pro 200 (we had paused it). If you have an existing plan you will keep the 20X multiplier for a bit and also receive a lot of additional credits because we know changes are hard even if it means that everyone will get more in the end.
1
431
Definitely
middot is the new emdash of AI smells
328
Hi, Tomorrow we are re-opening the Pro $200 subscriptions to new subscribers, but together with it we are also changing how we calculate the usage for it. In effect, if you do the math, it will net out at half the dollar in API spend compared to the old Pro $200 plan. Now that it's said, let me explain why this is happening and why you will still get more work done than if you were on the Pro $200 subscription one month ago. (a) We didn't want to compromise in other ways and are committing to not reintroducing the 5h limit, so that you can fully use the weekly usage when you want. (b) On the subscription, we guarantee that over time you always get more work done and with an increasing level of quality. This means that you will continue to get more value per dollar spent as a result of models getting more efficient and us passing down the improvements in the form of API price reductions. (c) We don't want to put an incentive on ourselves to artificially inflate the API list prices to make it look like you are getting a lot (and workaround it through discounts, etc). Instead we want to continue to both rapidly reduce prices and increase capabilities of models on the API. This week we introduced GPT-6 Sol and GPT-6 Luna at 50% of their previous price. Over time, we see prices go low enough that it makes sense for most to buy usage as needed without there being a significant gap between what you get in a subscription and what you get in the API for a dollar spent. (d) Tomorrow, we are adding more things to the subscription that won't draw on the usage, I won't reveal what that is yet. I wanted to be transparent before all the big announcements tomorrow. Lots of new exciting things are coming to the subscriptions that will make it super compelling, but I wanted to make sure to share this change ahead of time so you can all understand it before we shower you with good news. Codexingly, Tibo
67
358
28
10,857
746,188
Very doubtful that the real reason was "safety" as OpenAI has never cared about that. Most likely real reasons are: - Their benchmarks show that the model loses to Opus/Sonnet 5.5 - Really bad feedback and benchmarks of the previous 6.0 Sol which seemed to be smaller model meant to improve their profits instead of model performance and that lost to older 5.6 sol in almost all tests. Only good thing is that as more and more people are migrating to claude code codex is a lot faster now😅
1
373