<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"><channel><description>Senior Data Scientist | http://dziezyc.ai | Same handle on 𝕏</description><link>https://bsky.app/profile/dziezycmaciej.bsky.social</link><title>@dziezycmaciej.bsky.social - Maciej Dzieżyc</title><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3mqqpg4xvp22d</link><description>Moja polemika do wywiadu Marcina Giełzaka w podkaście Dwie Lewe Ręce z @kurasinski.bsky.social odbiła się lekkim echem, więc uznałem, że czas odświeżyć konto tutaj 😉 O tym, że polskie LMMy istnieją i że nie czeka nas (taka) apokalipsa. https://www.linkedin.com/pulse/o-tym-%C5%BCe-polskie-lmmy-istniej%C4%85-i-nie-czeka-nas-taka-polemika-dzie%C5%BCyc-twm5f/</description><pubDate>16 Jul 2026 07:25 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3mqqpg4xvp22d</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lwdvketnnc27</link><description>While coding with Claude vs GPT-5, it feels like they have inverse reward functions: Claude maximizes tokens (over-verbose responses/code), GPT-5 minimizes them (terse brevity). Both miss the optimal sweet spot IMHO.</description><pubDate>14 Aug 2025 08:15 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lwdvketnnc27</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lvzumr63ec2u</link><description>As another data point to costs savings and enshittification of ChatGPT/GPT-5 - first time I saw ChatGPT misspelling a Polish word (&#34;Akclimatyzacja&#34;)</description><pubDate>10 Aug 2025 08:32 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lvzumr63ec2u</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lvs5x5mjks2c</link><description>Is it just me, or does @github.com Copilot in @vscode.dev on WSL do weird things like editing files by executing sed in terminal because it can&#39;t do it the usual way?</description><pubDate>07 Aug 2025 06:57 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lvs5x5mjks2c</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3luuaixpmbc2s</link><description>🧐 Maybe making rich people richer is the only thing allowed? 😂</description><pubDate>26 Jul 2025 09:23 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3luuaixpmbc2s</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lumrde6ihc2f</link><description>If you&#39;re wondering how to create your own end-to-end encrypted (E2EE) backup using open-source tools, I wrote an article based on my experience 😅 medium.com/p/25b76fc30c00&#xA;https://medium.com/p/25b76fc30c00</description><pubDate>23 Jul 2025 10:03 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lumrde6ihc2f</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lltvfuwg4c2w</link><description>How did LLMs perform hours after the 2025 US Math Olympiad&#39;s problems were released? Well, they sucked, scoring 5% on average. Who would have thought? &#xA;&#xA;https://arxiv.org/abs/2503.21934v1&#xA;&#xA;cc @edzitron.com</description><pubDate>02 Apr 2025 17:18 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lltvfuwg4c2w</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3ligpyw6vuc2y</link><description>Looking for AI conferences worldwide? 👀 Here&#39;s a list of 20+ of them: https://medium.com/data-science-collective/20-applied-ai-conferences-in-2025-9e38bc942c34</description><pubDate>18 Feb 2025 07:21 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3ligpyw6vuc2y</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lhmb5ncbac2z</link><description>Days 3-4 with GitHub Copilot:&#xA;🚫 Struggled with data sources.&#xA;🚫 Misunderstood CSV headers.&#xA;🚫 Hallucinated SQL queries, although static suggestions were fine.&#xA;🚫 VS Code crashed a few times (OOM).&#xA;✅ Saved ~2 hours, finished an iteration today instead of Monday.&#xA;&#xA;[contains quote post or other embedded content]</description><pubDate>07 Feb 2025 18:46 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lhmb5ncbac2z</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lhgwh5yc4k2l</link><description>Day 2 with Github Copilot: Mostly Claude. It removed features I didn’t ask it to, struggled with splitting code into files, and had issues optimising dataset sampling while maintaining uniform distribution. Still like it, but definitely more issues as I went deeper.&#xA;&#xA;[contains quote post or other embedded content]</description><pubDate>05 Feb 2025 15:51 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lhgwh5yc4k2l</guid></item><item><link>https://bsky.app/profile/dziezycmaciej.bsky.social/post/3lhckhz5qqc2f</link><description>Today, I revisited GitHub Copilot after nearly a year away from it, and honestly, I was kind of blown away. I started working on a simple proof-of-concept Machine Learning project, and what would’ve normally taken me 1-3 days, I finished in just 4 hours. So far, Sonnet has been the most impressive.</description><pubDate>03 Feb 2025 22:06 +0000</pubDate><guid isPermaLink="false">at://did:plc:amsls2a5fobxrcsd2i4bknrf/app.bsky.feed.post/3lhckhz5qqc2f</guid></item></channel></rss>