ToKra

[François Chollet @fchollet]

The jobs of the future will require high adaptability and creativity, focusing on complex problem framing rather than repetitive execution or specialized skills

[CohenTheBarbarian @barbarian_cohen]

TGIF ;)

[devleader @DevLeaderCa]

I had one problem, so I solved it with enterprise Java.
…
..
.
Now I have an AbstractProblemFactoryFactory.

[Rahul @sairahul1]

As an AI Engineer. Please learn

>Harness engineering, not just prompt engineering

>Context engineering, not just long prompts

>Prompt caching vs. semantic caching tradeoffs

>KV cache management, eviction, reuse, and memory pressure at scale

>Prefill vs. decode latency and why they optimize differently

>Continuous batching, paged attention, and throughput optimization

>Speculative decoding vs. quantization vs. distillation tradeoffs

>INT8, INT4, FP8, AWQ, GPTQ, and when quantization hurts quality

>Structured output failures, schema validation, repair loops, and fallback chains

>Function calling reliability, tool contracts, argument validation, and idempotency

>Agent guardrails, loop budgets, tool budgets, and termination conditions

>Model routing, graceful fallback logic, and degraded-mode UX

>RAG architecture: chunking, embeddings, hybrid search, reranking, and freshness

>Retrieval evals: recall, precision, grounding, attribution, and citation quality

>Evals: golden sets, regression tests, adversarial tests, LLM-as-judge, and human evals

>LLM observability as a first-class discipline: traces, spans, tokens, latency, errors, and drift

>Cost attribution per feature, workflow, tenant, and user journey not just per model

>Safety engineering: prompt injection defense, data leakage prevention, and permission boundaries

>Multi-tenant isolation, cache safety, and cross-user context contamination prevention

>Fine-tuning vs. in-context learning vs. RAG vs. distillation and when each is the wrong tool

>Latency, quality, cost, and reliability tradeoffs across the full inference stack

>Production failure modes: hallucinated tool calls, malformed JSON, stale retrieval, runaway agents, and silent eval regressions

[vicki @vboykis]

Recommended reading if you feel you’ve lost your way , or if you’re having more fun than ever

[Marek Rašovský @Marek_Rasovsky]

Tohle se @SeznamZpravy povedlo. Tolik k paměti pamětníků.

[Richard Campbell @richcampbell]

MCP for Beginners - https://t.co/QC8a3ZV3Sz

[NEXTA @nexta_tv]

🔄 Ford just rehired 350 engineers it fired for AI

The company admitted AI couldn’t replace them.

Now the same engineers who were cut are back — this time to train new staff and actually make AI work.

Turns out replacing humans with AI wasn’t as easy as they thought 💀

[Thorsten Ball @thorstenball]

There's a lesson here for everybody who thinks that "I'll just use a cheaper model for certain tasks"

It's *very* hard to know in advance how smart a model has to be to do a task.

If it's not smart enough, it will likely keep trying, using cheaper but more tokens.

[John Carmack @ID_AA_Carmack]

AI may move to directly generating binary code, but I suspect there are still advantages to reasoning in a different representation.

Textual code is a flattening of an abstract syntax tree, and while LLMs produce tokens linearly, the prior context is only linearly connected by the relationship of the position embeddings, so I wonder if they could work more effectively if the position embeddings directly represented tree structures.

Code could be “parsed” into the context instead of directly entered into it.

[Lex Fridman @lexfridman]

Here's my conversation with Anthony Kaldellis about the deep history of the Roman Empire in the west and the east (the Byzantine Empire). This was a truly fascinating conversation with a lot of wisdom for the modern world and for the future of human civilization. The Roman state lasted over 2,200 years. If we want to understand human nature, the modern world, and how humanity can flourish, it is valuable to study history, especially the history of why societies survive and why they collapse.

It's here on X in full and is up everywhere else (see comment).

Timestamps:
0:00 - Episode highlight
1:24 - Introduction
1:51 - The Roman Empire and the Byzantine Empire
5:49 - 2,200 Years of Roman History
26:12 - Power, violence, and civil war
47:27 - Edict of Caracalla
1:00:23 - Crisis of the Third Century
1:14:52 - Constantine and the new Roman Empire
1:26:53 - Christianity in the Roman Empire
1:52:21 - Fall of the Western Roman Empire
2:05:17 - Eunuchs, Taxes, and Power
2:30:24 - Emperor Justinian and wars of conquest
2:47:26 - The Arab conquests
3:07:01 - Why the Roman empire survived so long
3:33:08 - Lessons from history

[Nádraží Vítkovice stan account @meh_srdce]

Přátelé kamarádi i letos sbírám nějaké místa v ČR kam se podívat. Čím větší bizár tím lépe! A čím méně známé, tím ještě líp! Předem díky za všechny tipy.

[François Chollet @fchollet]

Cross-agent feedback loops are incredibly effective -- for a reason. Check out what @leon2mcp and team at @Bloome_im are building in this space: https://t.co/9YeLjBNjsk

Bloome lets you pull Claude, ChatGPT, Gemini, and human teammates into a single shared workspace. The best feature is how your agents check each other's work. One drafts, another critiques, and another catches missing details. Human teammates can work in the same thread to keep the agents on target.

Having all your models and human coworkers in one shared context is wildly effective

[Microsoft Research @MSFTResearch]

AI agents can't remember past conversations. They must constantly reload or retrieve context, which grows less efficient as tasks get longer and more complex. Memora solves this with a scalable memory system separating what’s stored from how it's retrieved: https://t.co/XZfOqbA705

[History With Jacob @HistoryWJacob]

There was a small unit in World War II that operated deep behind Japanese lines more than a hundred times and never lost a single man.

They were called the Alamo Scouts.

Formed in late 1943 by General Walter Krueger, these men were hand-picked for one job: go where no one else could and come back with intelligence.

More than seven hundred volunteers tried out. He kept just 138. In teams of six or seven they slipped through the jungles of New Guinea and the Philippines, living for weeks inside enemy territory, mapping Japanese positions and sending back the intelligence that shaped entire invasions.

They did not just watch. On multiple missions they went in and brought people out. In New Guinea they freed dozens of captives the Japanese were holding.

Then came their most impressive mission:

American prisoners were starving in a camp at Cabanatuan, twenty-five miles inside Japanese-held Luzon. Before the Rangers could raid it, someone had to know exactly what was waiting. Two Alamo Scouts dressed as farm workers and set up a hidden observation post just a few hundred yards from the Japanese guards. For days they counted men and mapped the ground inch by inch.

The raid that followed freed more than five hundred Allied prisoners in a single night.

In total the Alamo Scouts ran 106 missions behind enemy lines across 1,482 days of sustained operations. Not one man was killed. Not one was captured.

The men who made it possible got almost none of the credit. Their work was secret, and secret men cannot be celebrated. When the war ended the Army no longer needed them, and the unit was quietly disbanded and forgotten.

The finest small-unit record of the war belonged to men most Americans have never heard of.

Now you have.

[Jiří Bigas @mrspockcz]

Simulátor pražského metra Back in Service, jehož vývoj vede Dominik Vojta, se dočkal dalšího velkého updatu. Vůbec poprvé se můžeme vydat mimo běžnou trasu linky A a vyjedeme na povrch, protože se hra rozšířila o Depo Hostivař.

[Gergely Orosz @GergelyOrosz]

I talked with a few folks inside Anthropic and I am starting to understand what @karpathy is saying (and what lots of people are misunderstanding)

It's not about Slack, but about a cloud AI, hooked up to ALL internal company systems, that "just works." THIS is the breakthrough

[.NET @dotnet]

Most devs over‑optimize the model. 🤖
Tweaking prompts & swapping models ≠ better outcomes.
Giving AI the right data + context = better reasoning + results.

Today's One Dev Question explores building the pipeline, not just the prompt.
🎥 https://t.co/9ENSs10itS

[.NET @dotnet]

.NET AI Community Standup: Multi-Agent Apps with Aspire + MAF https://t.co/vr78T7fwMU

[François Chollet @fchollet]

The best way to understand a complex system is via edge cases and failure modes, because they define the contour of the system.