Technology

Clearly the whole drama with Pentagon making a big deal of showing that they're trying to force AI companies to build autonomous AI killing machines and spy on citizens is completely manufactured.

Anthropic was always going to comply, and the goal is to just create a marketing campaign them as heroically resisting. All the media has been running the story of a plucky Anthropic defying US military to defend ethical AI and protect humanity.

https://archive.ph/W7KcJ

cross-posted from: https://news.abolish.capital/post/31311

Drop Site is a reader-funded, independent news outlet. Without your support, we can’t operate. Please consider becoming a paid subscriber or making a 501(c)(3) tax-deductible donation today.

Subscribe now

U.S. Army personnel monitor screens displaying maps and imagery of the Gaza Strip during a media tour inside the Civil-Military Coordination Center (CMCC) on November 20, 2025 in Kiryat Gat, Israel. The screen (above left) shows the use of Palantir’s Gaia application, billed as a tool to “bring the battlefield into view.” Photo by Amir Levy/Getty Images.

Palantir Technologies has a permanent desk at the U.S.-led Civil Military Coordination Center (CMCC) headquarters in southern Israel, three sources from the diplomatic community inside the CMCC told Drop Site News. According to the sources, the artificial intelligence data analytics giant is providing the technological architecture for tracking the delivery and distribution of aid to Gaza.

The presence of Palantir and other corporations—along with recent changes banning non-profits unwilling to give data to Israeli authorities—is creating a situation in which the delivery of aid is taking a backseat to the pursuit of profit, investment, and the training of AI products, experts say.

“The United Nations already has a humanitarian architecture in place to step in during crises, abiding by humanitarian principles and grounded in international law,” UN Special Rapporteur for the occupied Palestinian territory FrancescaAlbanese told Drop Site. “This profit-driven parallel system involving companies like Palantir, already linked to Israel’s unlawful conduct, can only be regarded as a monstrosity.”

The CMCC was established by U.S. Central Command (CENTCOM) in October, one week after the so-called ceasefire went into effect in Gaza to “monitor implementation of the ceasefire” and “help facilitate the flow of humanitarian, logistical, and security assistance from international counterparts into Gaza.” Last week, at the inaugural summit of the Board of Peace in Washington, D.C., Major General Jasper Jeffers—who was tapped in January to lead the International Stabilization Force in Gaza—announced that the CMCC will serve as the Board of Peace’s operational headquarters.

According to the sources, a representative from Palantir sits in the CMCC operations room where aid convoys and distributions inside Gaza are monitored through drone surveillance. The representative integrates convoy and distribution-related data into Palantir’s systems, the sources said.

Palantir did not respond to an inquiry from Drop Site on its role in the CMCC or in aid distribution in Gaza. Founded in 2003 by billionaire Peter Thiel with the help of investments from the CIA’s venture capital arm In-Q-Tel, Palantir is known for its work with government agencies, including the U.S. military and Immigration and Customs Enforcement (ICE).

In January 2024, three months into Israel’s war on Gaza, Palantir announced it had entered into a “strategic partnership” with the Israeli military for “war-related missions.” The company’s board meeting that month in Tel Aviv was held “in solidarity” with Israel, Bloomberg reported. Palantir did not disclose what technologies would be provided to Israel but a year earlier the company introduced its Artificial Intelligence Platform (AIP) to help militaries rapidly analyze and identify bombing targets. The company’s technology has been described by a Palantir executive as a way of “optimizing the kill chain.” Palantir’s software has also been used by the Israeli military in several raids in Gaza, according to a biography of its CEO, “The Philosopher in the Valley: Alex Karp, Palantir and the Rise of the Surveillance State” by Michael Steinberger.

In a June 2025 report to the UN Human Rights Council, Albanese found “reasonable grounds to believe Palantir has provided automatic predictive policing technology, core defence infrastructure for rapid and scaled-up construction and deployment of military software, and its Artificial Intelligence Platform, which allows real-time battlefield data integration for automated decision-making.”

The use of Palantir to track aid deliveries to Gaza is of particular concern to observers. “The distinction between death by drone and delivery of aid is being evaporated while we all sit around the same table,” a source from the diplomatic community who attends CMCC sessions told Drop Site.

Palantir’s two main platforms are Gotham and Foundry. “Gotham’s targeting offering supports soldiers with an Al-powered kill chain, seamlessly and responsibly integrating target identification and target effector pairing,” according to the company’s website. Foundry is Palantir’s platform for supply chain management and is billed as a way to “bridge siloed planning and execution processes, optimize inventory management, and help build supply chain resilience for economic and geopolitical uncertainty.”

Palantir does not operate its systems in isolated silos. According to the company’s own documentation, “Palantir AIP and Foundry are designed to interoperate with the full range of data, logic, AI, workflow, and security systems.” A feature called “Type Mapping” allows data entered into the civilian Foundry system to be instantly synchronized and queried by the military’s Gotham platform.

This means that, in theory, information that is being gathered at the CMCC—including from participating governments the UN and NGOs regarding the type of aid being distributed, its distribution locations and systems, and truck convoy routes—could be seamlessly pulled into Gotham’s AI targeting matrix. The same software logic used to track aid at the CMCC could be used to optimize and accelerate lethal airstrikes.

There is no information available as to whether Gotham and Foundry are the specific products being used to track aid, but publicly available photographic evidence indicates that Palantir’s Gaia—a platform referred to on their website as a tool to “bring the battlefield into view”—is being deployed at the CMCC.

In an interview with Drop Site on the role of Palantir in Gaza, the economist Yanis Varoufakis, the former Greek Finance Minister and a former member of Greece’s parliament, described an encounter he had with a Palantir representative who had explained to him the benefit the company gained from Gaza. “He was saying that ‘as the bombs fell we were having a party,’” Varoufakis said. According to Varoufakis, the Palantir representative explained how the chaos of intense violence in a high-density urban area like Gaza generates substantial data for training their AI models on how humans respond under stress. “The more bombardment and havoc, the better the training,” Varoufakis said.

“It’s one thing to say that companies like Lockheed Martin make money selling F35s to the Israelis,” he said. “That has been a time-honored way that the military industrial complex has benefited from war and genocide and war crimes.” He continued, “This is the first instance in history where it is the suffering of a people being subjected to genocide and bombing—the suffering itself—which is adding to the capital of a company which then uses that capital to produce commodities to sell elsewhere.”

Palantir operates on Oracle’s cloud infrastructure, led by Larry Ellison—a major donor to the Israeli military who also funds the Tony Blair Institute, which has itself consulted on governance mechanisms for Gaza.

Drop Site News is reader-supported. Consider becoming a free or paid subscriber.

The growing use of Palantir and other private sector companies in Gaza comes as the non-profit sector is being systematically squeezed out. As of March 1, 2026, Israel will ban dozens of aid groups from operating in Gaza, as well as the West Bank and East Jerusalem, under new registration rules, including prominent NGOs such as Doctors Without Borders, the Norwegian Refugee Council, Oxfam, and Medical Aid for Palestinians.

The new measures require aid groups to register the names and contact information of employees and to provide details about their funding and operations to Israeli authorities. The aid groups said in a joint statement this week that “the demand to transfer personal data raises acute security and legal risks. It exposes national staff to potential retaliation and undermines established data protection and confidentiality safeguards.”

“NGOs are being pushed out of Gaza because aid delivered by humanitarian organizations is based on need and is provided to people wherever they are located,” said an aid worker who spoke to Drop Site on condition of anonymity. “This doesn’t match the vision of the ‘New Gaza’ where Palestinians will need to be displaced again into the zones where reconstruction will be permitted and their access to aid will be controlled through screening.”

Not all NGOs are being pushed out of Gaza. Others—on a list of registered and approved organizations—are expanding their role alongside the private sector. These include Christian groups like Samaritan’s Purse and GAiN, both of whom were involved in the Gaza Humanitarian Foundation (GHF), and who sources from the diplomatic community have recently seen gathering in a “prayer circle” at the CMCC.

These approved NGOs, alongside private firms coordinated through the CMCC like Palantir, stand ready to take over the distribution of aid in Gaza.

“Aid in Gaza has been stripped to bare survival and its future delivery appears to be faith based, profit driven, militarized and certainly not to be delivered by anyone that dares to speak out about what Palestinians are being subjected to,” said a senior aid worker who spoke to Drop Site on condition of anonymity.

Gaza’s experience with private delivery mechanisms has been catastrophic. In May 2025, the U.S. and Israeli-backed GHF was contracted to distribute aid in the enclave. During the four and half months the GHF operated in Gaza, more than 2,600 Palestinians seeking food were killed and over 19,000 wounded by Israeli forces or security contractors at or near aid distribution sites.

The former headquarters of the GHF, a large warehouse-style building in Kiryat Gat, is now the headquarters of the CMCC.

As aid groups are being banned, U.S. military contractors are also filling the vacuum. According to sources from the diplomatic community who attend the CMCC, the physical presence of Safe Reach Solutions (SRS), a U.S. military contractor that provided security for the GHF, has recently expanded at the center of the facility, with SRS officials taking up more prominent seating space on the operations floor. The company’s representatives now sit behind name tags in seating that had previously been reserved for UN agencies, the sources said.

SRS did not respond to an inquiry from Drop Site about its role at the CMCC or in Gaza aid distribution.

“Given the precedent of the GHF, which turned aid delivery into a killing machine,” Albanese told Drop Site, “and the grave violations of international law embedded in the so-called peace plan—first and foremost the negation of the Palestinian right to self-determination—the risk that companies and states involved in the CMCC may be complicit in, or even direct perpetrators of, international crimes is real.”

Arkel International, a longtime U.S. military contractor, has also had representatives at CMCC briefing sessions, according to the sources inside the center. Arkel recruited drivers from Serbia and Georgia to drive supplies into Gaza for the GHF in 2025, according to Haaretz. At the time, Arkel was represented in Israel by businessman Hezi Bezalel, who has served as honorary consul of Rwanda in Israel.

The re-emergence of GHF-linked companies, alongside the digital capacity to monitor and monetize the surviving population in Gaza, is now converging with the construction of a new physical infrastructure spearheaded by giant real estate conglomerates.

At last week’s Board of Peace meeting, the reconstruction of Gaza was positioned as a massive financial “unlocking” of a distressed asset. Figures like Yakir Gabay, who built a real estate empire in Germany, envisioned the coastline transformed into a “Mediterranean Riviera” featuring 200 hotels and artificial islands. Marc Rowan, a billionaire investor and the CEO of Apollo Global Management, framed the project as the consolidation of Gaza’s “productive assets” into a “unified structure.”

A significant addition to the CMCC’s corporate roster is Terra Firma Capital Partners, which sources confirmed now maintains a permanent presence at the CMCC. Founded by British financier Guy Hands, the firm brings experience in managing massive-scale residential assets. Terra Firma has links to the New Labour era, specifically through Lord John Birt, Tony Blair’s former strategy director, who worked for the company after he left government.

“The genocide is entering a new phase. After the destruction of Gaza and the erasure of entire family lines, powerful states are now deciding the fate of the survivors without ever listening to their voices,” Albanese said.

“If Gaza is not to become a capitalist techno-dystopia, the time to act is now. States and corporations supporting this emerging infrastructure must be stopped, and held accountable. There is no time to lose.”

Leave a comment

Share

From Drop Site News via This RSS Feed.

Regular LoRA training is basically a standard gradient descent optimization loop where you have to curate a dataset, run backpropagation, and slowly update the low-rank matrices over many steps. It is computationally expensive and tedious every single time you want to teach the model a new trick or feed it a new document.

What Sakana AI built with Doc-to-LoRA completely bypasses that repetitive training loop at deployment time by introducing a hypernetwork. They shifted the massive computational burden upfront through a meta-training phase where a separate neural network actually learns how to predict the correct LoRA weights directly from an input document or task description.

Once that hypernetwork is trained, generating a new LoRA adapter only takes a single sub-second forward pass instead of a full fine-tuning run. You just feed a document into the frozen base model to get its token activations, and the hypernetwork instantly spits out the custom LoRA weights. This is incredibly effective for solving the long-term memory bottleneck in large language models.

Instead of shoving a massive document into the context window for every single query, which completely eats up your VRAM and spikes latency, you permanently internalize that knowledge into a tiny adapter footprint of under fifty megabytes. They also designed a clever chunking mechanism that processes the document in small segments and concatenates the resulting adapters. This allows the model to perfectly recall information from documents that are tens of thousands of tokens longer than its actual native context limit. It essentially turns a slow and expensive engineering pipeline into a cheap and instant forward pass.

source code https://github.com/SakanaAI/Doc-to-LoRA

I thought others would be interested in this network stack called reticulum. I really love the idea of building a decentralized internet that allows wide communication without reliance on big tech. This could be useful in something like a natural disaster or a situation where a government is interfering with communications (esp since it can be portable). It is not fast enough for something like streaming, but can be used for text based communication and is encrypted. It does have some existing applications for both desktop and android. I'm going to try it out and will report back on how it works

Reticulum

Machine learning community has been stuck on the autoregressive bottleneck for years, but a new paper shows that it's possible to use diffusion models to work on discrete at scale. The researchers trained two coding focused models named Mercury Coder Mini and Small that completely shatter the current speed and quality tradeoff.

Independent evaluations had the Mini model hitting an absurd throughput of 1109 tokens per second on H100 GPUs while the Small model reaches 737 tokens per second. They are essentially outperforming existing speed optimized frontier models by up to ten times in throughput without sacrificing coding capabilities. On practical benchmarks and human evaluations like Copilot Arena the Mini tied for second place in quality against huge models like GPT-4o while maintaining an average latency of just 25 ms. Their model matched the performance of established speed optimized models like Claude 3.5 Haiku and Gemini 2.0 Flash Lite across multiple programming languages while decoding exponentially faster.

The advantage of diffusion relative to classical autoregressive models stems from its ability to perform parallel generation which greatly improves speed. Standard language models are chained to a sequential decoding process where they must generate an answer exactly one token at a time. Mercury abandons this sequential bottleneck entirely by training a Transformer model to predict multiple tokens in parallel. The model starts with a sequence of pure random noise and applies a denoising process that iteratively refines all tokens simultaneously in a coarse to fine manner until the final text emerges. Because the generation happens in parallel rather than sequentially the algorithm achieves a significantly higher arithmetic intensity that fully saturates modern GPU architectures. The team paired this parallel decoding capability with a custom inference engine featuring dynamic batching and specialized kernels to squeeze out maximum hardware utilization.

Saved me for the second time in 6 months, holy shit.

Flatpak problems seem to keep happening on my machine, and this time I just couldn't do anything with flatpak at all. Apps wouldn't launch, terminal commands wouldn't work. No browser, communications, email, nothing since everything is now on flathub.

Thankfully I had backups set up. One is timeshift, which incrementally saves my system and boot files, and the other is the distro's own Backup tool, which is a bit slow but saves my entire home directory every few days.

I was able to restore the missing files from the backup, and everything works again. I'm now going through the issue to figure out how exactly it happened.

Last time flatpak decided to just delete its own PGP keys so it couldn't communicate with itself anymore. A timeshift backup fixed that - I have since set up hourly backups over 24 hours, since they're incremental they don't take up a lot of space and they delete the older ones automatically. Default home backup app takes a bit more space, I'm going to look at replacing it with something a bit leaner and faster if possible.

Still took a bit of time, but we're talking maybe 15 minutes instead of a potential multi-day or even week-long headache.

But if you're switching to linux, do yourself a favor and get a spare 1TB drive that you will use exclusively for backups. It's not optional, this will literally save you.

If you want to do cool shit you can even set up a NAS in your basement and backup on there, otherwise honestly even just an external drive you plug in is enough, as long as you remember to do at least daily backups.

This paper is honestly one of the most creative takes on LLM reasoning I’ve seen in a while. The team at ByteDance basically argues that we should view Long Chain-of-Thought as a macromolecular structure with internal forces that hold the logic together. They found that when we try to teach a model to reason by simply distilling keywords from a teacher, it fails because it’s like trying to build a protein by looking at a photo of it rather than understanding the atomic bonds.

Their Molecular Structure of Thought hypothesis breaks reasoning down into three specific bond types that behave similarly to their chemical counterparts. Deep reasoning acts like covalent bonds, forming the rigid primary backbone where each logical step must strictly justify the next. Self-reflection functions like hydrogen bonds, creating folding patterns where the model looks back 100 steps to audit an earlier premise, which keeps it from hallucinating. Finally, you have self-exploration acting like van der Waals forces, these are low-commitment bridges that let the model probe different ideas without getting stuck in a rigid path too early.

They found that most synthetic reasoning data is actually trash because it lacks this distribution. They proved that models don't actually learn the keywords themselves, but the characteristic reasoning behaviors those keywords represent. In one experiment, they replaced keywords like wait with arbitrary synonyms or removed them entirely, and the models still learned the reasoning structure just fine. It turns out that building these stable thought molecules is what creates the basis for Long CoT, as opposed to just mimicking a specific vibe or prompt format.

They built MOLE-SYN to address the problem. Instead of just copying teacher outputs, it uses a distribution transfer graph to walk through four behavioral states to synthesize traces that have the correct bond profile from the start. Their approach makes reinforcement learning much more stable because the model starts with a balanced skeleton instead of a bunch of fragmented logic. The paper challenges the whole more data is better mindset to argue that it's the geometry of the information flow that really matters.

A blog post I found in response to Cory Doctorow taking a pro-LLM stance in a recent post of his.

https://archive.ph/6tjt3