Linas Petkevicius

President @Lithuania_AI | EAIForum Board | Director Institute of CS @VU_LT | SR @AIML_VU | Editor @NAMC_VU | Deep Learning |

[Linas Petkevicius @dxlino]

85 overall CAM, Target Man. built different, deployed different.

get scouted → https://t.co/nRgSRXCJGW #GitFut

[Linas Petkevicius @dxlino]

🎉 Huge congrats to Vilnius University (@VU_LT) Software Engineering Class of '26!

Proud of all the top bachelor & master grads — especially the talented students I supervised on cutting-edge AI theses:
- Vision-Language Models for Lithuanian medical data,
- reducing hallucinations,
- agentic AI for healthcare, and more.

Top of the class: https://t.co/R14Xg1emCC

Shoutout to the program chairs making it happen! 📷#SoftwareEngineering #VU #AI

[Linas Petkevicius @dxlino]

Wrapping up an epic Deep Learning semester with the Class of 2026! 🚀

- 120+ students mastered Deep learning + hands-on VLMs/SLMs,
- built CV & generative AI projects, and
- fine-tuned models for Lithuanian (low-resource language boost).

Huge props to top bachelors & masters performers + amazing TAs! Proud of the next gen of AI talent. 📷

Top students: https://t.co/seGjtyKgkf

#DeepLearning #VLM #ClassOf2026

[Linas Petkevicius @dxlino]

🚀Postdoctoral Fellowship in AI for Remote Sensing at Vilnius University (Lithuania). Work on deep learning, VLM's, multi-task imagery, and AutoResearch at @AIML_VU . 24 months, €3634 gross/month, start as early as Sep 2026. Apply by 22 June 2026 → https://t.co/YogA4i9sLW

[Linas Petkevicius @dxlino]

https://t.co/yzCTCW74eX

[Linas Petkevicius @dxlino]

https://t.co/wE2h5k9yg2

[Linas Petkevicius @dxlino]

2024 AI -> 2025 AI Agents -> 2026 Physical AI was it seems the most dominant AI keyword in this year Davos.

Among interisting AI news in Davos:
- Uncertainty in access to capital (if pulled back) could slow down the growth of AI ecosystem (@robheyvaert);
- Early results of AI agents raise the expectations of first evidence of AI impact of jobs market (@DarioAmodei);
- Vocaly raised risk to limit AI as legal person status (@Yoshua_Bengio);
- Concerns of too fast AI migration in our lives (history claims, were never dealt with it well, @harari_yuval);

and many more topics apart, geopolitics was discussed this year.

[Linas Petkevicius @dxlino]

https://t.co/H9B03qX7W1

[Linas Petkevicius @dxlino]

https://t.co/7mChMaRKGH

[Linas Petkevicius @dxlino]

@elonmusk Maybe then it could parse bigger number to primes than 21 = 7*3 :)
Great quantum lobbying efforts, to say enemies will decript you messages unless you fund quantum :)
While technology is not proved to solve tasks any real world tasks.

[Linas Petkevicius @dxlino]

@briedi3 V8 Porsche?

[Linas Petkevicius @dxlino]

Day 3 of Davos 2025: The coordinated steps to lock the ROI of AI have been taken

While users underestimate LLM's performance of free versions, the decitions (500B initial investment in AI infrastructure, restrictions of chips export, The Organization for Economic Co-operation and Development (OECD) Global Tax Deal (Global Tax Deal), Declaring a National Energy Emergency, Unleashing American Energy), making unified steps to ensure that within two years the competitive advantage of AI models would be concentrated in one country. When buissiness finally moving to adapt AI at scale, the two year window until next iteration of LLM's/AI models placing stong safeguards to ensure that the ROI of AI will be locked in one place.

Another interesting notes:
- Dario Amodei (Anthropic) expects continued growth of LLMs quality for next two years to reach superhuman AI
- Jonas Andrulis (Aleph Alpha) anounced about released Tokenizer-Free (T-Free) architecture
- Aidan Gomez (Cohere) expressed expectations for continuation of growth industry

[Linas Petkevicius @dxlino]

Day 2 of Davos 2025: Expectations of economics of AI - capitalization of data centers and infrastructure up, capitalization of new SaaS up, capitalization of software creation shrinks.

The LLM's, AI agents will shrink the capitalization of software creation and integration costs up to magnitude of order. The capitalization of data centers will continue to grow next decade and be limited by energy infrastructure (capacity available producers and stability of grid); The "one person companies" will flood the market with SaaS solutions (with automated integrations); However, the capabilities of LLM's and AI agents will increase the productivity of software developers and integrators, so the capitalization of software creation will shrink the cost software products and services due to the competition, this will lead to shrinking of the capitalization of software creation.

Another interesting notes:
- Samuel Bonamigo, of Databricks, expects that it's just beginning of 4th industrial revolution, just starting, since we just starting to understand the data at scale;
- Kiril Evtimov of G42 raised four pillow of energy issues: electricity and grid reliability, connectivity, insufficient infrastructure and lack of chip supply.
- A Politico debates on "Does need a kill switch" heat up, with amazing debates, on AI systems and it's risks;
- The Economist's organized debate on AI in media lead that even old processes could have 10-40% increase of productivity with AI, where new processes could have even bigger increase of productivity;
- Rephrasing Jonas Andriulis, CEO of Aleph-Alpha we in new world conqueration time, where we will be either users of AI products which might biased and not transparent or we will be owners of our future with our sovereignty and transparency.

[Linas Petkevicius @dxlino]

The results of tokeniser free model adaptation for different languages (like Finnish) are booming in comparison to fine/tunning by @Aleph__Alpha and @JonasAndrulis

[Linas Petkevicius @dxlino]

Day 1 of Davos 2025: Democratisation of AI access opened the pandora box and the wild wild west began.

The ChatGPT and all LLM enabling CS knowledge access all around the globe for everyone with a laptop. The competitiveness in years to come will become truly global, there are no more borders for creating IT/software solutions. Improvise, adapt and overcome or you will be left behind. The new era of AI is here.

Another interesting notes:
- The next AI leap would not be based on GenAI models, according to Yann Lecun. Problem solving and reasoning would be the next frontier, which aim to solve the problems without teaching how to solve them.
- Paul Alivisatos, addressing the new challenge in education: what does it mean to be educated in the age of AI?
- Bolor-Erdene Battsengel, addressed the new inclusivity issue, among lack of digital and social access, the next big challenge on widening the gap between the ones with access to AI and the ones without.
- And ofcouse inauguration. The positive expectations from blockchain folks and AI industry from less regulatory expectations in the US.

[Linas Petkevicius @dxlino]

@darkooo Kai nesisekė, sezono pradžia ir antras pagal mažiausiai minučių gaunantis first round pick’as
Aišku labai gaila nes Dunovanas tai dar nuo Sabonio laikų rookies nepasitikėjo, nors Čikaga tai į lotery vėl eis bet ne bando duot minutės veteranams, ech

[Linas Petkevicius @dxlino]

@darkooo Turbūt piratinis stream, nukiptas logo ir kraštai, nei pristatymo, nei plaukimo. baisu :)

[Linas Petkevicius @dxlino]

@edvvis Gal pranzūziškai buvo, ir vietoje plaukimo transliacijos nusipirko tinklinį, nes kam tv3 čia tie tautiečiai

[Linas Petkevicius @dxlino]

@briedi3 O jei neturėti Iron Dome gi šakės https://t.co/ZB6KDABcZ4

[Linas Petkevicius @dxlino]

@Tomce544 čia toks apgaulingas, ten gan advanced knyga tho