<rss version="2.0" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Hacker News: jakozaur</title><link>https://news.ycombinator.com/user?id=jakozaur</link><description>Hacker News RSS</description><docs>https://hnrss.org/</docs><generator>hnrss v2.1.1</generator><lastBuildDate>Wed, 29 Jul 2026 01:49:22 +0000</lastBuildDate><atom:link href="https://hnrss.org/user?id=jakozaur" rel="self" type="application/rss+xml"></atom:link><item><title><![CDATA[Tokenflation: When "Hi" triggers 33 tool calls]]></title><description><![CDATA[
<p>Article URL: <a href="https://quesma.com/blog/tokenflation-when-hi-triggers-33-tool-calls/">https://quesma.com/blog/tokenflation-when-hi-triggers-33-tool-calls/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=49055561">https://news.ycombinator.com/item?id=49055561</a></p>
<p>Points: 6</p>
<p># Comments: 1</p>
]]></description><pubDate>Sun, 26 Jul 2026 07:15:16 +0000</pubDate><link>https://quesma.com/blog/tokenflation-when-hi-triggers-33-tool-calls/</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=49055561</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=49055561</guid></item><item><title><![CDATA[New comment by jakozaur in "Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k"]]></title><description><![CDATA[
<p>This isn’t limited to large system prompts. Coding-agent harnesses are also becoming more aggressive about using tools, even for trivial requests. In our tests, prompts such as “Hey” or “commit” sometimes triggered 30+ tool calls:<p><a href="https://quesma.com/blog/the-true-cost-of-saying-hi-to-an-ai-agent/" rel="nofollow">https://quesma.com/blog/the-true-cost-of-saying-hi-to-an-ai-...</a><p>Tokenflation seems very real: the number of tokens consumed by simple tasks keeps increasing.</p>
]]></description><pubDate>Sun, 12 Jul 2026 18:55:45 +0000</pubDate><link>https://news.ycombinator.com/item?id=48883533</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48883533</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48883533</guid></item><item><title><![CDATA[New comment by jakozaur in "Claude Code May–July 2026 weekly limits promotion"]]></title><description><![CDATA[
<p>Fable is also extended by another week till July 19:
<a href="https://support.claude.com/en/articles/15424964-claude-fable-5-promotional-access" rel="nofollow">https://support.claude.com/en/articles/15424964-claude-fable...</a></p>
]]></description><pubDate>Sun, 12 Jul 2026 18:15:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=48883176</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48883176</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48883176</guid></item><item><title><![CDATA[Make Europe Cool Again]]></title><description><![CDATA[
<p>Article URL: <a href="https://jacek.migdal.pl/2026/07/07/make-europe-cool-again">https://jacek.migdal.pl/2026/07/07/make-europe-cool-again</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=48861061">https://news.ycombinator.com/item?id=48861061</a></p>
<p>Points: 6</p>
<p># Comments: 0</p>
]]></description><pubDate>Fri, 10 Jul 2026 15:12:31 +0000</pubDate><link>https://jacek.migdal.pl/2026/07/07/make-europe-cool-again</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48861061</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48861061</guid></item><item><title><![CDATA[New comment by jakozaur in "Show HN: Smart model routing directly in Claude, Codex and Cursor"]]></title><description><![CDATA[
<p>It's rather hard to do at the proxy level with agentic coding, such as Claude Code or similar. These are long-chained sessions of tool use that heavily rely on prompt caching. Changing mid-flight is costly.<p>It looks like much more context is required to decide on the best model (e.g., summarizing logs might use a cheap model, whereas you likely want Opus/Mythos/GPT 5.6 to debug multithreading logic). In an agentic system, a decision about the model may be embedded in the decision to orchestrate the model.</p>
]]></description><pubDate>Fri, 26 Jun 2026 18:30:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=48690211</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48690211</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48690211</guid></item><item><title><![CDATA[New comment by jakozaur in "Poland is now among the 20 largest economies"]]></title><description><![CDATA[
<p>Poland's first partly free election was on 4 June 1989, preceded by the roundtable negotiations.<p>The protests in Czechoslovakia came later, called the Velvet Revolution, from 17 to 28 November 1989. In June 1990, Czechoslovakia held its first democratic elections, a year after Poland.<p>Poland paved the way for the whole of central and eastern Europe. The Round Table produced the negotiated-exit template that Hungary built on in its own talks that summer, and that Czechoslovakia, East Germany, and the Baltics drew on as their regimes fell within months.<p>And it did so from the deepest macroeconomic crisis of any of the satellite states: hyperinflation running into the hundreds of percent by late 1989, an unresolved sovereign default from 1981, and chronic shortages.<p>Since then Poland has converged fastest of any of them. From a low base it has climbed to the upper-middle of central and eastern Europe by GDP per capita PPP, overtaken Hungary, and is now closing on Czechia and Slovenia.</p>
]]></description><pubDate>Fri, 08 May 2026 16:15:29 +0000</pubDate><link>https://news.ycombinator.com/item?id=48065176</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48065176</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48065176</guid></item><item><title><![CDATA[New comment by jakozaur in "Poland is now among the 20 largest economies"]]></title><description><![CDATA[
<p>The story is longer: Poland was the first country to make a remarkable peaceful transition from a bankrupt, failed Soviet satellite state. The shock therapy, plus NATO and EU aspirations, paved the way.<p>It is a story of a country that made a lot of the right decisions along the way. Managed to keep consistent high growth, not a pony trick or boom/bust mode.<p>Poland should be a role model for many other countries.<p>Recommend a book: <a href="https://www.amazon.com/Europes-Growth-Champion-Insights-Economic/dp/0198789343" rel="nofollow">https://www.amazon.com/Europes-Growth-Champion-Insights-Econ...</a><p>And Noah's blog post:
<a href="https://www.noahpinion.blog/p/the-polandmalaysia-model" rel="nofollow">https://www.noahpinion.blog/p/the-polandmalaysia-model</a></p>
]]></description><pubDate>Fri, 08 May 2026 15:52:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=48064871</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48064871</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48064871</guid></item><item><title><![CDATA[New comment by jakozaur in "Heat pump sales rise across Europe"]]></title><description><![CDATA[
<p>A heat pump could win as the best HVAC technology, though a better drilling for ground-sourced ones. Just a shallow drilling (up to 100m) that works in retrofit mode, such as drilling from the basement, would be a great upgrade:<p>- No outdoor unit that looks awful in many settings<p>- works well, even in the coldest winter, without a spike in electricity usage, COP 5<p>- very reliable with long durability<p>- super quiet, no ambient noise<p>- 20% more efficient<p>Currently, drilling is very disruptive in retrofits, but there is progress in compact techniques that might change the equation.<p>Disclaimer: angel investor in <a href="https://www.flexdrill.at/" rel="nofollow">https://www.flexdrill.at/</a></p>
]]></description><pubDate>Mon, 04 May 2026 19:03:07 +0000</pubDate><link>https://news.ycombinator.com/item?id=48013371</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=48013371</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=48013371</guid></item><item><title><![CDATA[New comment by jakozaur in "$500 GPU outperforms Claude Sonnet on coding benchmarks"]]></title><description><![CDATA[
<p>Build systems are tested by CompileBench (Quesma's benchmark).<p>Disclaimer: I'm the founder.</p>
]]></description><pubDate>Fri, 27 Mar 2026 11:16:59 +0000</pubDate><link>https://news.ycombinator.com/item?id=47541300</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47541300</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47541300</guid></item><item><title><![CDATA[New comment by jakozaur in "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"]]></title><description><![CDATA[
<p>I can provide trajectories. Though probably we are not going to publish them this time. This would need some extra safeguards.<p>Email me. The address is in profile.</p>
]]></description><pubDate>Sun, 22 Feb 2026 17:22:01 +0000</pubDate><link>https://news.ycombinator.com/item?id=47112771</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47112771</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47112771</guid></item><item><title><![CDATA[New comment by jakozaur in "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"]]></title><description><![CDATA[
<p>As I do eval and training data sets for living, in niche skills, you can find plenty of surprises.<p>The code is open-source; you can run it yourself using Harbor Framework:<p>git clone git@github.com:QuesmaOrg/BinaryAudit.git<p>export OPENROUTER_API_KEY=...<p>harbor run 
 --path tasks 
 --task-name lighttpd-* 
 --agent terminus-2 
 --model openrouter/anthropic/claude-opus-4.6 
 --model openrouter/google/gemini-3-pro-preview 
 --model openrouter/openai/gpt-5.2 
 --n-attempts 3<p>Please open PR if you find something interesting, though our domain experts spend fair amount of time looking at trajectories.</p>
]]></description><pubDate>Sun, 22 Feb 2026 16:21:16 +0000</pubDate><link>https://news.ycombinator.com/item?id=47112255</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47112255</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47112255</guid></item><item><title><![CDATA[New comment by jakozaur in "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"]]></title><description><![CDATA[
<p>So many models refuse to do that due to alignment and safety concerns. So cross-model comparison doesn't make sense. We do, however, require proof (such as providing a location in binary) that is hard to game. So the model not only has to say there is a backdoor, but also point out the location.<p>Your approach, however, makes a lot of sense if you are ready to have your own custom or fine-tuned model.</p>
]]></description><pubDate>Sun, 22 Feb 2026 16:17:18 +0000</pubDate><link>https://news.ycombinator.com/item?id=47112212</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47112212</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47112212</guid></item><item><title><![CDATA[New comment by jakozaur in "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"]]></title><description><![CDATA[
<p>Oh, nice find... We end up using PyGhidra, but the models waste some cycles because of bad ergonomics. Perhaps your cli would be easier.<p>Still, Ghidra's most painful limitation was extremely slow time with Go Lang. We had to exclude that example from the benchmark.</p>
]]></description><pubDate>Sun, 22 Feb 2026 16:11:03 +0000</pubDate><link>https://news.ycombinator.com/item?id=47112155</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47112155</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47112155</guid></item><item><title><![CDATA[New comment by jakozaur in "We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them"]]></title><description><![CDATA[
<p>See direct benchmark link: <a href="https://quesma.com/benchmarks/binaryaudit/" rel="nofollow">https://quesma.com/benchmarks/binaryaudit/</a><p>Open-source GitHub: <a href="https://github.com/QuesmaOrg/BinaryAudit" rel="nofollow">https://github.com/QuesmaOrg/BinaryAudit</a></p>
]]></description><pubDate>Sun, 22 Feb 2026 14:56:38 +0000</pubDate><link>https://news.ycombinator.com/item?id=47111486</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47111486</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47111486</guid></item><item><title><![CDATA[We hid backdoors in ~40MB binaries and asked AI + Ghidra to find them]]></title><description><![CDATA[
<p>Article URL: <a href="https://quesma.com/blog/introducing-binaryaudit/">https://quesma.com/blog/introducing-binaryaudit/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47111440">https://news.ycombinator.com/item?id=47111440</a></p>
<p>Points: 245</p>
<p># Comments: 98</p>
]]></description><pubDate>Sun, 22 Feb 2026 14:50:44 +0000</pubDate><link>https://quesma.com/blog/introducing-binaryaudit/</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47111440</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47111440</guid></item><item><title><![CDATA[New comment by jakozaur in "Ghidra by NSA"]]></title><description><![CDATA[
<p>Funny thing, AI is not that terrible at using Ghidra. We released a benchmark on that and hopefully models will improve:
<a href="https://quesma.com/blog/introducing-binaryaudit/" rel="nofollow">https://quesma.com/blog/introducing-binaryaudit/</a></p>
]]></description><pubDate>Mon, 16 Feb 2026 15:44:43 +0000</pubDate><link>https://news.ycombinator.com/item?id=47036445</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=47036445</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=47036445</guid></item><item><title><![CDATA[New comment by jakozaur in "Show HN: Ghidra MCP Server – 110 tools for AI-assisted reverse engineering"]]></title><description><![CDATA[
<p>Funny coincidence, I'm working on a benchmark showcasing AI capabilities in binary analysis.<p>Actually, AI has huge potential for superhuman capabilities in reverse engineering. This is an extremely tedious job with low productivity. Currently reserved, primarily when there is no other option (e.g., malware analysis). AI can make binary analysis go mainstream for proactive audits to secure against supply-chain attacks.</p>
]]></description><pubDate>Wed, 04 Feb 2026 09:00:00 +0000</pubDate><link>https://news.ycombinator.com/item?id=46883334</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=46883334</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46883334</guid></item><item><title><![CDATA[New comment by jakozaur in "Pg_tracing: Distributed Tracing for PostgreSQL"]]></title><description><![CDATA[
<p>Great idea. Currently, people have to rely on client-side spans in OpenTelemetry. However, it would be awesome if we could get spans for slow SQL queries, along with explanations.</p>
]]></description><pubDate>Sun, 01 Feb 2026 10:01:13 +0000</pubDate><link>https://news.ycombinator.com/item?id=46844970</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=46844970</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46844970</guid></item><item><title><![CDATA[New comment by jakozaur in "Benchmarking OpenTelemetry: Can AI trace your failed login?"]]></title><description><![CDATA[
<p>In this benchmark, micro-services are really small, ~300 lines, and sometimes just two of them. More realistic tasks (large codebases, more microservices) would have a lower success rate.</p>
]]></description><pubDate>Thu, 29 Jan 2026 16:53:29 +0000</pubDate><link>https://news.ycombinator.com/item?id=46812771</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=46812771</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46812771</guid></item><item><title><![CDATA[OTelBench: Can AI instrument OpenTelemetry?]]></title><description><![CDATA[
<p>Article URL: <a href="https://quesma.com/benchmarks/otel/">https://quesma.com/benchmarks/otel/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=46766032">https://news.ycombinator.com/item?id=46766032</a></p>
<p>Points: 5</p>
<p># Comments: 1</p>
]]></description><pubDate>Mon, 26 Jan 2026 14:27:07 +0000</pubDate><link>https://quesma.com/benchmarks/otel/</link><dc:creator>jakozaur</dc:creator><comments>https://news.ycombinator.com/item?id=46766032</comments><guid isPermaLink="false">https://news.ycombinator.com/item?id=46766032</guid></item></channel></rss>