Sercan Külcü

Sercan Külcü

Assistant Professor, Computer Engineering, Giresun University

Blog Posts

Google Releases Gemini 1.5

3 minute read

Published:

Google announced Gemini 1.5 Pro on February 15, 2024 in a research preview that made it available to select developers through Google AI Studio and the Gemini API. The headline capability was a context window of up to one million tokens — roughly 750,000 words, equivalent to approximately 30,000 lines of code, 11 hours of audio, 1 hour of video, or several complete novels in a single request. For comparison, GPT-4 Turbo (announced November 6, 2023) offered a 128,000-token context, and Claude 2.1 (November 2023) offered 200,000 tokens; Gemini 1.5’s one-million-token window was five to eight times larger than any commercially available alternative at the time. Google demonstrated the capability with tasks including analyzing an entire 402-page technical manual to answer detailed questions, understanding a full-length movie provided as video frames, and performing few-shot in-context learning from hundreds of examples rather than the dozens that fit in previous context windows. The model also supported multimodal inputs — text, images, audio, and video — within the same context window, enabling queries like “at what timestamp in this hour-long lecture does the professor make an error in this equation?” Read more

Apple Vision Pro Reviews Bring Spatial Computing Into Focus

3 minute read

Published:

Apple opened Vision Pro preorders on January 19, 2024, selling out the first wave in minutes; early pre-order confirmations showed delivery dates in late February. Review embargoes lifted February 1, 2024, one day before the retail launch on February 2. The $3,499 starting price (with prescription lens inserts adding $149–$299) resulted in a device that journalists handled as a premium technology artifact rather than a consumer product, and reviewer consensus reflected that tension: extraordinary engineering at a price and form factor that limited its audience. Read more

Google Releases Gemini

3 minute read

Published:

Google introduced Gemini on December 6, 2023, as a natively multimodal model family in three sizes: Nano (running on-device in Pixel 8 Pro phones), Pro (replacing Bard as Google’s conversational AI), and Ultra (the largest version, subject to further safety testing before release). Unlike earlier Google models that added multimodal capability by aligning separate text and vision encoders, Gemini was trained from the start on interleaved sequences of text, images, audio, and video, allowing reasoning that crossed modalities within a single pass. Read more

Java Superstars

4 minute read

Published:

Java 21 arrived in September 2023 as the latest long-term support release, and it is a significant one. Virtual threads — the main deliverable of Project Loom — are now production-ready. Structured concurrency and scoped values are in preview. Pattern matching for switch is finalized. Record patterns work in instanceof checks. After years of shorter improvement cycles since Java 9, the platform has changed considerably. The people shaping this evolution are worth understanding, but so is the community structure that makes Java’s development sustainable. Read more

OpenAI’s Leadership Crisis Shakes the AI Industry

2 minute read

Published:

On November 17, 2023, OpenAI’s board — composed of Ilya Sutskever (co-founder and Chief Scientist), Adam D’Angelo (Quora CEO), Tasha McCaulay (FTX Future Fund), and Helen Toner (Georgetown CSET) — fired CEO Sam Altman in a surprise announcement, stating he “was not consistently candid in his communications with the board.” Greg Brockman, President and co-founder, resigned in protest within hours. The board named CTO Mira Murati interim CEO, then replaced her with Emmett Shear (co-founder of Twitch) days later. The timing came 11 days after OpenAI’s first developer conference (DevDay, November 6), where GPT-4 Turbo and the GPT Store had been announced, with ChatGPT at approximately 100 million weekly active users. Read more

Staying Engaged In Computer Science Classes

5 minute read

Published:

Computer science classes can become difficult for a reason that has nothing to do with intelligence: the subject moves from concrete to abstract unusually fast. A student learns to write a loop on Monday and hears about time complexity, memory models, and formal proofs before the month ends. It is easy to lose the thread between what you can see on the screen and what the teacher is now talking about. Read more

Ubuntu 23.10 Shows the New Desktop Installer Direction

3 minute read

Published:

Ubuntu 23.10 “Mantic Minotaur” was released on October 12, 2023 as a nine-month interim release between the Ubuntu 23.04 (April 2023) and Ubuntu 24.04 LTS (April 2024) releases. The most significant infrastructure change was the default deployment of Canonical’s new Flutter-based desktop installer, replacing the Ubiquity installer that had been the standard Ubuntu desktop installation tool since Ubuntu 5.10 “Breezy Badger” in October 2005. Ubiquity was built on Python and GTK, and while it had received updates throughout its 18-year life, its architecture made it difficult to share code between the desktop installer, the Ubuntu Server installer (Subiquity, introduced in Ubuntu 18.04), and the Ubuntu Core IoT/embedded installer. The new installer, written in Dart using the Flutter framework (the same cross-platform UI toolkit Google uses for mobile apps), shared the Subiquity backend used for server installations and presented a redesigned step-by-step interface with cleaner visuals, accessibility improvements, and a new partitioning UI. The choice of Flutter was notable: it was a cross-platform framework primarily associated with mobile and web development, not traditional Linux desktop software, reflecting Canonical’s assessment that Flutter’s widget consistency across platforms made it the most practical choice for sharing a single codebase across the desktop, server, and embedded installer variants. Other desktop changes included GNOME 45 (released September 20, 2023), which featured a new Quick Settings panel replacing the old system menu, an updated Activities overview with a workspace indicator in the top bar, and a breaking change to GNOME Shell extension APIs that required all existing extensions to be updated for GNOME 45 compatibility — a recurring pattern in GNOME releases that frustrated extension maintainers. Read more

Apple Introduces the A17 Pro

2 minute read

Published:

Apple introduced the A17 Pro on September 12, 2023, exclusively in the iPhone 15 Pro ($999) and Pro Max ($1,199) — the first time Apple gave a chip a “Pro” designation to distinguish it from the A16 Bionic continuing in the iPhone 15 and 15 Plus. The A17 Pro was built on TSMC’s first-generation 3nm process (N3B, using EUV lithography), containing 19 billion transistors — up from the A16 Bionic’s 16 billion transistors on TSMC’s 4nm node. The A17 Pro’s 6-core CPU delivered 2 high-performance cores (at approximately 3.78 GHz) and 4 efficiency cores; Apple claimed 10% faster CPU performance than A16 while improving power efficiency at similar performance levels. Read more

Why Choose Computer Engineering in 2023?

3 minute read

Published:

In 2023, generative artificial intelligence has changed the public conversation about computing. Tools that can produce text, images, and code have made many people ask what software work will look like in the future. Read more

LK-99 Claims Trigger Intense Computational Analysis

2 minute read

Published:

On July 22, 2023, two preprint papers appeared on arXiv from Sukbae Lee and colleagues at the Quantum Energy Research Center in Seoul, claiming that a lead-apatite compound doped with copper — chemical formula Pb₉Cu(PO₄)₆O, dubbed LK-99 — exhibited superconductivity at room temperature and ambient pressure. A video showing a small pellet partially levitating above a magnet (a demonstration of the Meissner effect, in which superconductors expel magnetic fields) spread on social media within days. If true, the claim would be the most significant physics discovery in decades: room-temperature superconductors would transform power transmission, computing, and MRI machines by eliminating the need for liquid helium or liquid nitrogen cooling. Read more

Meta Releases Llama 2

2 minute read

Published:

Meta released Llama 2 on July 18, 2023 in partnership with Microsoft, distributing model weights through Hugging Face, llama.meta.com, and Microsoft Azure. The release included three sizes — 7B, 13B, and 70B parameters — in both base (pretrained) and Chat (RLHF-aligned) variants, under a custom commercial license that permitted use by organizations with fewer than 700 million monthly active users without additional licensing, and required a separate agreement from Meta for larger deployments. This was a significant change from the first LLaMA (February 2023), which had been released for research-only use and whose weights leaked via torrent within a week. Read more

The Gaming PC I Would Build in 2023

3 minute read

Published:

If I were building a gaming PC on July 1, 2023, this is the machine I would actually want. I am not simply choosing the most expensive part in every category. The priority is gaming performance, then reliability, cooling and enough headroom to keep the system useful for years. Read more

Apple Introduces the Vision Pro

3 minute read

Published:

Apple introduced Vision Pro at WWDC on June 5, 2023 as a “spatial computer” starting at $3,499. It went on sale February 2, 2024 in the US. The headset contained two micro-OLED displays (23 million pixels total across both eyes, higher than a 4K TV viewed from a typical distance), driven by an M2 chip for application processing and a new R1 chip for real-time sensor fusion. Twelve cameras, six microphones, and five sensors including LiDAR and a TrueDepth camera fed data continuously to the R1, which processed and delivered images to the displays within 12 milliseconds — a latency Apple stated was below human perceptual thresholds for lag between head movement and visual update. Read more

Google Announces the PaLM 2 Model

4 minute read

Published:

Google announced PaLM 2 at Google I/O on May 10, 2023, as the successor to PaLM (Pathways Language Model, April 2022, 540 billion parameters). PaLM 2 was offered in four sizes — Gecko (deployable on mobile devices), Otter, Bison, and Unicorn (the largest, with parameter count undisclosed) — allowing Google to serve the same model family from on-device inference to large cloud deployments. Google’s technical report described PaLM 2 as trained on a significantly broader multilingual corpus (100+ languages and programming content spanning 20+ programming languages) than PaLM, with a smaller model-to-data ratio (a “chinchilla-optimal” approach, citing Hoffmann et al. 2022’s finding that smaller models trained on more data outperform larger models on less data at the same compute budget). Read more

SpaceX Starship Completes Its First Integrated Flight Test

3 minute read

Published:

SpaceX launched Starship’s first integrated flight test (IFT-1) on April 20, 2023 at 8:33 AM CDT from Starbase in Boca Chica, Texas. The fully stacked vehicle — Super Heavy booster (B7) plus Starship upper stage (S24) — stood approximately 120 meters (394 feet) tall, making it the largest rocket ever flown. Super Heavy’s 33 Raptor 2 engines produced approximately 74.4 MN (16.7 million pounds-force) of liftoff thrust, surpassing the Saturn V (35.1 MN) and NASA’s Space Launch System Block 1 (39.1 MN) to make it the most powerful rocket ever successfully launched from the pad. The vehicle successfully cleared the launch tower and rose to approximately 38 km altitude over the Gulf of Mexico before the mission ended with an unplanned vehicle breakup. During the ascent, several Raptor engines failed to ignite or shut down early — approximately 6 of 33 engines were observed non-functional during flight — though the vehicle was designed to fly with 3–4 engine failures. Stage separation (planned as a “hot staging” maneuver, with the upper stage’s engines igniting before booster engine shutdown) did not execute as planned; the vehicle began rotating uncontrollably approximately 2:50 into the flight. The flight termination system (FTS) was triggered by SpaceX’s flight safety officer approximately 4 minutes into the flight as the vehicle descended uncontrolled over the Gulf. Read more

Gordon Moore Has Passed Away At The Age Of 94

4 minute read

Published:

Gordon Moore died on March 24, 2023, at the age of 94. His name is inseparable from one of the most consequential observations in the history of technology: Moore’s Law. The observation shaped not only the semiconductor industry but the software written to run on it, and the expectations of everyone who used computers for half a century. Read more

GPT-4 Is Released

2 minute read

Published:

OpenAI released GPT-4 on March 14, 2023, making it available to ChatGPT Plus subscribers ($20/month) and via the API to developers on a waitlist. Microsoft had already integrated GPT-4 into Bing Chat (launched February 7, 2023) before the formal public release — an early signal that enterprise integration was ahead of consumer access. OpenAI published a GPT-4 Technical Report the following day, disclosing extensive benchmark results but deliberately omitting parameter count, training compute, dataset composition, and architecture specifics, citing “competitive and safety reasons.” This represented a notable break from the research transparency of GPT-2 (2019) and GPT-3 (2020), both of which had technical papers with full methodology. On standard academic benchmarks, GPT-4 scored in the 90th percentile on the bar exam (GPT-3.5 had scored the 10th percentile), in the 93rd percentile on the SAT reading/writing section, and 88th percentile on SAT math. On MMLU (57 academic subjects, 5-shot), GPT-4 scored 86.4% versus GPT-3.5’s 70%. The model launched with 8,192 token context (a 32K context version was available to select API partners), compared to GPT-3.5’s 4,096 tokens. Read more

Meta Releases LLaMA

3 minute read

Published:

Meta released LLaMA (Large Language Model Meta AI) on February 24, 2023, as a family of decoder-only transformer models at four sizes: 7B, 13B, 33B, and 65B parameters. Training used 1 to 1.4 trillion tokens drawn from Common Crawl (67%), GitHub, Wikipedia, books, ArXiv papers, and Stack Exchange — a much larger dataset than GPT-3’s 300 billion tokens. The 13B model matched or exceeded GPT-3 (175B parameters) on most standard benchmarks, demonstrating that smaller models trained longer on cleaner data could match much larger models trained on fewer tokens, a finding that echoed the “Chinchilla scaling laws” paper DeepMind had published in March 2022. Read more