Raspberry Pi killed more Russians than HIMARS, self-replicating AI, and a world without programmers

Good morning, friends! With you again your favorite mega show about artificial intelligence «SHOW ON AI» and its hosts: Oleksiy Minakov (AI apologist in Ukraine), Serhiy Kolesnychenko (AI realist, COO UCloud) and Stas Yurasov (CEO dev.ua). Despite the hoarse voices after celebrating the 5th anniversary of dev.ua, we gathered to give you the most powerful analytics at the intersection of philosophy, business and technology in 2026:)

In the first part of the issue, we analyze the main trend of Silicon Valley, which overshadowed AGI — the concept of RSI (Recursive Self-Improvement), or recursive self-improvement of models. How Anthropic forced AI to write 80% of its own code, why «model collapse» occurs and how governments are trying to hide new neural networks (Fable 5, Mythos 5) from foreigners? We discuss Ukraine’s AI sovereignty, the «agent state» and why the education system will have to be razed to the ground.

Special guest of the issue: Vladislav Piotrovsky, co-founder and CEO of DWARF Engineering. Uncensored about the most technological sphere of war:

• Have Ukrainian drones really turned into autonomous killer «terminators»?
• Why is Russia scaling technology faster and how are Wi-Fi modems and Raspberry Pi saving the day?
• Drone vs. drone warfare, laser and microwave weapons.
• The phenomenon of «kitchen» UAVs: why is assembling household items cooler than an assembly line factory?

In the finale — AI practice: we test Ukrainian models Mamai LM and Lap LLM on the Lapatonia Chat platform and create a real Hollywood movie, setting Stas Yurasov’s hand on fire and forcing Oleksiy’s avatar to hide in the toilet via Gemini Omni!

The release is sponsored by cloud technology provider UCloud. Reliable clouds and intelligent text AI agent Claudia, which will quickly unload your support!

Release timecodes:

00:00 — Teaser: Ukraine plans to produce 7 million drones in 2026. Swarm or autonomy?

01:14 — Start of the show «SHOW PO SHI»! Greetings from the hosts: Alexey Minakov, Sergey Kolesnichenko and Stas Yurasov.

01:51 — UCloud Sponsor Integration: Text AI agent Claudia saves support operators.

02:51 — What is RSI (Recursive Self-Improvement) and why is it more important than the concept of AGI?

03:31 — Silicon Valley Trend: Anthropic Study «When AI builds itself.»

04:36 — What is recursive self-improvement in simple words? Point improvements to the architecture.

05:33 — How can an AI teach another AI if it does not know the historical context (analogy with a socket and current).

06:16 — Multi-agent systems: how an «agent-as-a-judge» (LLM-as-a-Judge) validates the hypotheses of other bots.

07:35 — A chess-like competition of ideas. Why science needs RSI, and not all industries.

09:01 — AI as a basis for creativity: 1% talent and 99% automated AI «sweat of the bush.»

10:19 — Risk of loss of control and the problem of Model Collapse (degeneration of AI knowledge at the 10th iteration).

10:42 — Real facts: Anthropic creates over 80% of its code without humans using automated reviewers.

12:15 — How high-quality is AI code and why the role of programmers is rapidly decreasing.

12:54 — Will a human understand the Anthropic Claude 5 model, which is fully trained by another artificial intelligence?

14:03 — Andriy Karpaty’s project «Autoresearch» based on GPT-2: recreating the evolution of technologies.

15:33 — RSI as a complex system with positive feedback and automation of the operational cycle.

17:00 — AI’s time limit: why breakthrough «genius» models need weeks of autonomous thinking.

18:17 — Token economy: how users subconsciously shape the «character» and limits of different models.

19:13 — A real-world case study of corporate AI that offers daily improvements based on interaction.

20:09 — Who controls the Black Box? The risks of AI autonomy and the cost of error.

21:33 — Isolation precedent: why the US government banned foreigners from accessing the new Fable 5 and Mythos 5 models.

23:26 — The concept of AI sovereignty. Are we in danger of receiving «weaker» models in Ukraine?

24:31 — Megacorporations (Microsoft, Google) as a «state within a state» and their competition with institutions.

25:28 — Palantir’s prediction: OpenAI and Anthropic will be nationalized by the US government.

26:02 — G7 Summit in France: Zelensky and CEOs of leading AI companies at the same table.

27:02 — The conflict between Elon Musk (SpaceX) and NASA: who really writes the blueprints for the space industry.

28:32 — Why Ukraine needs its own AI sovereignty and the collection of national datasets.

29:32 — Why Ukrainian business is learning American models, but will soon switch to open-source China.

31:04 — How European regulations (AI Act) kill innovation and why Ukraine put them on pause.

32:39 — Are Ukrainian laws written using self-improvement methods? Criticism of lawmaking.

33:34 — The concept of the Agentic State. Transformation of the Ministry of Digital Affairs after Fedorov’s transfer to the Ministry of Defense.

35:37 — The Legal Department of the Ministry of Digital Affairs has automated 75% of its work and processing of applications through AI.

36:11 — What will happen if we spam the Verkhovna Rada with AI bills? Mariana Bezuglaya’s outburst in Rozenka.

38:00 — Why lawmaking and state control will never yield to AI: fear of losing statehood.

38:24 — Anthropic’s idea: to wash your own island to escape US regulations forever.

39:20 — Boeing 737 Max software bug and AI analogy with nuclear weapons escalation.

40:46 — How should the education system change in 2026? Why teach children if the world is non-linear.

42:15 — Intelligence explosion: release time for new models reduced to two weeks. Money flywheel and IPOs.

43:26 — What skills will be relevant in 10 years: critical thinking, flexibility, and soft skills.

44:15 — Presenters' conclusions: how Tetris at 9th speed prepares children for constant Reskilling (reprofiling).

46:13 — Experience of the conference in Ivano-Frankivsk: the problem is not in firing people, but in the speed of their retraining.

47:02 — The world of the future: part-time employment, shifting responsibility to AI, and more time for coffee.

48:54 — New trade infrastructure: access to GPU clusters instead of access to the sea and oceans.

49:55 — Factories without light and people in China. Shortage of robotics engineers in Ukraine.

51:04 — Are classical universities needed? Competition between private universities (UCU, KSE) and public ones.

52:14 — «SHI-GUEST» SECTION: Introduction by Vladislav Piotrovsky (CEO DWARF Engineering).

52:51 — Creating an OS for mass deployment of applications on UAVs and AI co-pilots for reaching the destination in case of loss of communication.

53:44 — The challenge of scaling technologies to factories and transferring code under strict security protocols.

54:59 — DWARF Engineering simulators for the military that copy the interface and save drones from crashing.

55:43 — Article in New Scientist about Ukrainian drones in «terminator» mode: the West’s view versus our pragmatism.

57:05 — What is true autonomy on the battlefield and why drones are now simply replacing artillery and missiles.

58:40 — The scaling problem: why the totalitarian regime in Russia stifles innovation faster, despite the lack of a free market.

1:00:41 — Creation of kill zones and autonomous patrolling of the contact line by Kakhanovsky’s drones.

1:01:43 — Combat in three planes. Why air combat automation is necessary due to the shortage of operators.

1:02:34 — Tactics of small Russian infantry groups and technical limitations of conventional FPV drones via radio channels.

1:04:10 — Is Swarm technology realistic? Bee choreography versus UAV rotation.

1:06:33 — Creating an autonomous drone is like writing Word. Creating a swarm is just adding italics.

1:07:07 — How field command views AI autonomy and why FPV pilots fear being transferred to infantry.

1:08:30 — Operation «Web» from the SBU (destruction of bombers in the Russian Federation using Computer Vision).

1:10:08 — The race of interceptor drones. The economic inefficiency of shooting down Shahids with $1,000,000 missiles.

1:11:59 — Radar networks, audio and video sensors in Ukraine: how to make a drone cheaper than a regular cartridge.

1:12:46 — Secret developments: laser and microwave weapons that fry enemy electronics from a distance.

1:13:49 — Is it still relevant to assemble drones in the kitchen? A decentralized network that cannot be destroyed by a missile.

1:15:50 — Logistical flexibility of manual assembly. When you need a factory, serial soldering, and a torque wrench.

1:17:29 — Mykhailo Fedorov as the new Minister of Defense: roundtables with Flash, Sternenko, and software codification.

1:19:32 — Enemy progress: autonomous attacks by Lancets without radio communication and the Yolka interceptor.

1:20:50 — How China helps the Russian Federation: supplying FPV cameras to official production and purchasing all optics a year in advance.

1:22:34 — The largest drone attack on Moscow and refineries: why the Russian air defense exposed its own infrastructure.

1:24:19 — A window of opportunity of six months for Ukraine. Why it is impossible to cover the area of ​​Russia with detection.

1:25:42 — What’s next? Interceptor detection and jamming systems, firearms on UAVs.

1:27:10 — Infrastructure and OS for gluing software with different hardware as the main game changer of war.

1:28:09 — 4-6 weeks of the technology life cycle on the front. A drone is not a weapon, but a platform for capabilities.

1:29:45 — Comparison of Ukraine and the USA: 7 million fast food drones versus expensive foreign MIL-STD standards.

1:31:17 — Why did the head of Rheinmetall contemptuously call Ukrainian UAVs «household kitchen appliances.»

1:32:32 — The bureaucracy of Western defense giants. How Denmark is the first to switch to short annual contracts for startups.

1:33:52 — Restrictions on arms exports to Ukraine: why we are sculpting deep-tech from Wi-Fi modems and Raspberry Pi.

1:35:42 — World experience in innovation: DARPA in the USA and the MAFAT organization (1.5 thousand PHD engineers) in Israel.

1:37:28 — Ground-based robotic systems (GROS): history of the volunteer project of the «Shablya» turret since 2016.

1:39:05 — Why controlling a drone on the ground is many times more difficult than flying in the air: SLAM technology and cross-country ability.

1:40:31 — Sensational capture of an occupier using a ground drone: how it works operationally.

1:41:10 — What is full autonomy for NRC? Robots that must learn to dig trenches and camouflage themselves.

1:43:00 — «SHI-PRACTICE» SECTION: We are testing the Lapatonia Chat platform and the domestic models Mamai LM and Lap LLM based on Gemma.

1:44:46 — Secure data processing inside Ukraine. Testing the new RADA model for sensitive issues about the war.

1:46:48 — Who created the Mamai model with 27 billion parameters: Anna Yakymenko’s team (Switzerland-Bulgaria).

1:48:22 — Why text models can’t draw and why business and the state need AI sovereignty.

1:49:50 — Video editing in Gemini Omni: spoken prompt «add fire to the palm and smoke» for Stas Yurasov.

1:51:19 — Gemini 10 second limit. Comparison with Runway and how the video effect works in real time.

1:53:14 — Creative trick with AI avatars: we create a video based on a text script «Alexey Minakov flees to the toilet from explosions.»

1:54:41 — Moment of truth: demonstration of Stas Yurasov’s finished video with fire and smoke accompanied by real sound.

1:55:24 — Demonstration of Alexey’s AI avatar with explosions and fireworks in the background of a bedroom.

1:56:16 — The show’s finale: calls to like, share the release, and wishes to engage in recursive self-improvement! Bye!