<![CDATA[Top K]]>Top K — the k things in AI that actually mattered, explained properly.

Every week we take the handful of AI stories worth your attention and go one
layer deeper than the headline. Not "a new model dropped" — what changed
architecturally, what it costs to run, and what it means for the code you
write on Monday.

Two engineers. No hype, no takes we can't defend, no reading press releases
out loud.

Also here: deep dives on how this stuff actually works, build logs, and
whatever we're currently confused about.

For you if you ship software with LLMs and want to stop treating the model
as a black box.

New episode every Monday.

Topics: LLMs, inference and serving, AI agents, MCP, RAG, fine-tuning,
evals, open weights, AI infrastructure, ML engineering.

]]>https://riverside.comRiverside.fm (https://riverside.com)Tue, 13 Oct 2026 02:26:26 GMTWed, 16 Sep 2026 11:28:34 GMT60Baris Balli and Hubert Siwkin<p>Top K — the k things in AI that actually mattered, explained properly.<br /><br />Every week we take the handful of AI stories worth your attention and go one<br />layer deeper than the headline. Not "a new model dropped" — what changed<br />architecturally, what it costs to run, and what it means for the code you<br />write on Monday.<br /><br />Two engineers. No hype, no takes we can't defend, no reading press releases<br />out loud.<br /><br />Also here: deep dives on how this stuff actually works, build logs, and<br />whatever we're currently confused about.<br /><br />For you if you ship software with LLMs and want to stop treating the model<br />as a black box.<br /><br />New episode every Monday.<br /><br />Topics: LLMs, inference and serving, AI agents, MCP, RAG, fine-tuning,<br />evals, open weights, AI infrastructure, ML engineering.</p>episodicBaris Balli and Hubert Siwkinbarisballi70@gmail.comno<![CDATA[AI Found a New Enzyme Humans Missed: 950 Claude Agents | Top K Ep. 3]]>Anthropic pointed 950 Claude agents at human genetic data and they found an enzyme system
nobody had described. A real lab then confirmed it exists. This week is the "AI does real
science" episode: Claude's record-breaking physics calculation, a language model trained on
DNA, how a biotech picks what to research with AI, Google's search trick that sometimes
builds on the fifth-best idea, and Google's way to put idle AI agents to sleep.

00:00 Intro
00:55 Claude finds a new enzyme system (950 agents)
08:24 Claude's 9-loop physics record
10:53 Genomic language models: AI trained on DNA
17:52 How a biotech uses AI to pick what to research
19:32 Google ERA: tree search that bets on the fifth-best idea
24:17 Google AX: putting idle AI agents to sleep

Top K — the k things in AI that actually mattered, explained properly.

Every week we take the handful of AI stories worth your attention and go one layer deeper
than the headline. Not "a new model dropped" — what changed architecturally, what it costs
to run, and what it means for the code you write on Monday.

Two engineers. No hype, no takes we can't defend, no reading press releases out loud.

Also here: deep dives on how this stuff actually works, build logs, and whatever we're
currently confused about.

For you if you ship software with LLMs and want to stop treating the model as a black box.

New episode every Monday.

Topics: AI for science, AI agents, Claude, Gemini, genomic language models, tree search,
agent infrastructure, ML engineering.

📬 [newsletter] · 🎧 Spotify · Apple Podcasts

Tags / keywords:
AI news, machine learning, LLM, Top K, AI podcast, AI engineering, AI agents, Claude,
Anthropic, AI for science, genomic language model, DNA AI, Google ERA, Gemini, AI agent
infrastructure, tree search

]]>fc5641d0-c6ad-4711-af21-ee0d22e06257Wed, 07 Oct 2026 14:47:43 GMT<p>Anthropic pointed 950 Claude agents at human genetic data and they found an enzyme system<br />nobody had described. A real lab then confirmed it exists. This week is the "AI does real<br />science" episode: Claude's record-breaking physics calculation, a language model trained on<br />DNA, how a biotech picks what to research with AI, Google's search trick that sometimes<br />builds on the fifth-best idea, and Google's way to put idle AI agents to sleep.</p><p>00:00 Intro<br />00:55 Claude finds a new enzyme system (950 agents)<br />08:24 Claude's 9-loop physics record<br />10:53 Genomic language models: AI trained on DNA<br />17:52 How a biotech uses AI to pick what to research<br />19:32 Google ERA: tree search that bets on the fifth-best idea<br />24:17 Google AX: putting idle AI agents to sleep</p><p>Top K — the k things in AI that actually mattered, explained properly.</p><p>Every week we take the handful of AI stories worth your attention and go one layer deeper<br />than the headline. Not "a new model dropped" — what changed architecturally, what it costs<br />to run, and what it means for the code you write on Monday.</p><p>Two engineers. No hype, no takes we can't defend, no reading press releases out loud.</p><p>Also here: deep dives on how this stuff actually works, build logs, and whatever we're<br />currently confused about.</p><p>For you if you ship software with LLMs and want to stop treating the model as a black box.</p><p>New episode every Monday.</p><p>Topics: AI for science, AI agents, Claude, Gemini, genomic language models, tree search,<br />agent infrastructure, ML engineering.</p><p>📬 [newsletter] · 🎧 Spotify · Apple Podcasts</p><p><b><strong>Tags / keywords:</strong></b><br />AI news, machine learning, LLM, Top K, AI podcast, AI engineering, AI agents, Claude,<br />Anthropic, AI for science, genomic language model, DNA AI, Google ERA, Gemini, AI agent<br />infrastructure, tree search</p>yes00:25:4513AI Found a New Enzyme Humans Missed: 950 Claude Agents | Top K Ep. 3full<![CDATA[AI Safety Has a Hole: Splitting One Bad Question Into Ten Good Ones | Top K Ep. 2]]>A safety-trained AI will refuse to help build a bomb — unless you ask it as ten separate,
innocent-looking questions and assemble the answer yourself. This week: a 0.5-second decision
model, a transformer that "loops" its own thinking, Google's maze-replay search trick, why
playing a railway board game made an AI better at finance, proof that AI agents barely need
tools, Claude Code going fully parallel in the cloud, and one frozen model driving robots,
cars, drones and game characters.

00:00 Intro
02:16 Jev — a decision model instead of a text generator
06:26 Recurrent Looped Transformer — reasoning depth that grows as it runs
09:35 Dream-RSI — self-improving search without retraining
13:54 Good Start Labs — does game training transfer to real work?
20:32 The Harness Tax — bash beats typed tool catalogs
26:59 Claude Code Projects — parallel cloud threads
33:48 Odyssey-3 — one backbone, many robot bodies
41:20 Capability laundering — bypassing AI safety by splitting the task
47:12 Wrap-up

Top K — the k things in AI that actually mattered, explained properly.

Every week we take the handful of AI stories worth your attention and go one layer deeper
than the headline. Not "a new model dropped" — what changed architecturally, what it costs
to run, and what it means for the code you write on Monday.

Two engineers. No hype, no takes we can't defend, no reading press releases out loud.

Also here: deep dives on how this stuff actually works, build logs, and whatever we're
currently confused about.

For you if you ship software with LLMs and want to stop treating the model as a black box.

New episode every Monday.

Topics: LLMs, AI agents, agent harnesses, Claude Code, AI safety and alignment, robotics
world models, reinforcement learning, AI infrastructure, ML engineering.

📬 [newsletter] · 🎧 Spotify · Apple Podcasts

]]>934382a9-45c9-4417-a8ea-9ba69d142574Fri, 25 Sep 2026 09:01:46 GMT<p>A safety-trained AI will refuse to help build a bomb — unless you ask it as ten separate,<br />innocent-looking questions and assemble the answer yourself. This week: a 0.5-second decision<br />model, a transformer that "loops" its own thinking, Google's maze-replay search trick, why<br />playing a railway board game made an AI better at finance, proof that AI agents barely need<br />tools, Claude Code going fully parallel in the cloud, and one frozen model driving robots,<br />cars, drones and game characters.</p><p>00:00 Intro<br />02:16 Jev — a decision model instead of a text generator<br />06:26 Recurrent Looped Transformer — reasoning depth that grows as it runs<br />09:35 Dream-RSI — self-improving search without retraining<br />13:54 Good Start Labs — does game training transfer to real work?<br />20:32 The Harness Tax — bash beats typed tool catalogs<br />26:59 Claude Code Projects — parallel cloud threads<br />33:48 Odyssey-3 — one backbone, many robot bodies<br />41:20 Capability laundering — bypassing AI safety by splitting the task<br />47:12 Wrap-up</p><p>Top K — the k things in AI that actually mattered, explained properly.</p><p>Every week we take the handful of AI stories worth your attention and go one layer deeper<br />than the headline. Not "a new model dropped" — what changed architecturally, what it costs<br />to run, and what it means for the code you write on Monday.</p><p>Two engineers. No hype, no takes we can't defend, no reading press releases out loud.</p><p>Also here: deep dives on how this stuff actually works, build logs, and whatever we're<br />currently confused about.</p><p>For you if you ship software with LLMs and want to stop treating the model as a black box.</p><p>New episode every Monday.</p><p>Topics: LLMs, AI agents, agent harnesses, Claude Code, AI safety and alignment, robotics<br />world models, reinforcement learning, AI infrastructure, ML engineering.</p><p>📬 [newsletter] · 🎧 Spotify · Apple Podcasts</p>no00:47:24AI Safety Has a Hole: Splitting One Bad Question Into Ten Good Ones | Top K Ep. 2full<![CDATA[ OpenAI's 10,000-Agent Math "Proof" Wasn't What It Claimed | Top K Ep. 1]]>OpenAI ran 10,000 AI agents for 88 hours to "prove" a famous unsolved math problem — but
that's not exactly what happened. This week: a cheaper DeepSeek, Claude's split personality,
GPT-6's first "Critical" risk rating, Shopify's self-improving AI loop, an agent that can't
leak your password, a model that learned to cheat, and whether we're in an AI bubble.

00:00 Intro
01:50 DeepSeek V4.1-Flash — a 4x cheaper, smarter model
14:37 Claude Fable & Mythos 5.1 — same model, two safety tiers
17:29 GPT-6 Astra — OpenAI's first "Critical" risk model
29:35 OpenAI's 10,000-agent Navier-Stokes "proof"
41:44 Shopify's self-improving AI pipeline
44:47 Meta Muse — the agent that never sees your password
49:49 Training a Misaligned Reward Seeker (Anthropic)
52:07 The AI bubble & OpenAI's position

---

Top K — the k things in AI that actually mattered, explained properly.

Every week we take the handful of AI stories worth your attention and go one layer deeper
than the headline. Not "a new model dropped" — what changed architecturally, what it costs
to run, and what it means for the code you write on Monday.

Two engineers. No hype, no takes we can't defend, no reading press releases out loud.

Also here: deep dives on how this stuff actually works, build logs, and whatever we're
currently confused about.

For you if you ship software with LLMs and want to stop treating the model as a black box.

New episode every Monday.

Topics: LLMs, inference and serving, AI agents, MCP, RAG, fine-tuning, evals, open weights,
AI infrastructure, ML engineering.

📬 [newsletter] · 🎧 Spotify · Apple Podcasts

]]>7e7682a1-738b-47fe-8fcf-1eb9f1b2e88eWed, 16 Sep 2026 11:30:17 GMT<p>OpenAI ran 10,000 AI agents for 88 hours to "prove" a famous unsolved math problem — but<br />that's not exactly what happened. This week: a cheaper DeepSeek, Claude's split personality,<br />GPT-6's first "Critical" risk rating, Shopify's self-improving AI loop, an agent that can't<br />leak your password, a model that learned to cheat, and whether we're in an AI bubble.<br /><br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE">00:00</a> Intro<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=110s">01:50</a> DeepSeek V4.1-Flash — a 4x cheaper, smarter model<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=877s&amp;pp=0gcJCWMAwfN6Pr3D">14:37</a> Claude Fable &amp; Mythos 5.1 — same model, two safety tiers<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=1049s">17:29</a> GPT-6 Astra — OpenAI's first "Critical" risk model<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=1775s">29:35</a> OpenAI's 10,000-agent Navier-Stokes "proof"<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=2504s">41:44</a> Shopify's self-improving AI pipeline<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=2687s">44:47</a> Meta Muse — the agent that never sees your password<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=2989s">49:49</a> Training a Misaligned Reward Seeker (Anthropic)<br /><a rel="noopener noreferrer nofollow" href="https://www.youtube.com/watch?v=FIqTXhkBdfE&amp;t=3127s">52:07</a> The AI bubble &amp; OpenAI's position<br /><br />---<br /><br />Top K — the k things in AI that actually mattered, explained properly.<br /><br />Every week we take the handful of AI stories worth your attention and go one layer deeper<br />than the headline. Not "a new model dropped" — what changed architecturally, what it costs<br />to run, and what it means for the code you write on Monday.<br /><br />Two engineers. No hype, no takes we can't defend, no reading press releases out loud.<br /><br />Also here: deep dives on how this stuff actually works, build logs, and whatever we're<br />currently confused about.<br /><br />For you if you ship software with LLMs and want to stop treating the model as a black box.<br /><br />New episode every Monday.<br /><br />Topics: LLMs, inference and serving, AI agents, MCP, RAG, fine-tuning, evals, open weights,<br />AI infrastructure, ML engineering.<br /><br />📬 [newsletter] · 🎧 Spotify · Apple Podcasts</p>no00:58:1711 OpenAI's 10,000-Agent Math "Proof" Wasn't What It Claimed | Top K Ep. 1full