Nvidia's attempt to salvage its reputation over DLSS 5 took a notably softer turn this week, with CEO Jensen Huang admitting he understands where gamers and developers are coming from. In a podcast interview with Lex Fridman published on 23 March, Huang said: "I don't love AI slop myself" and acknowledged that he is "empathetic towards what they're thinking."
This marks a sharp pivot from his earlier stance at the GTC 2026 conference, where he flatly told critics they were "completely wrong" about the technology.
However, Huang's more diplomatic tone masks deeper problems that technical details and developer reactions have since exposed. The fundamental tension remains unresolved: does DLSS 5 operate at the geometry level, as Huang claims, or does it function as a 2D image filter, as technical details suggest?
According to reporting from Kotaku, Nvidia's own "GeForce Evangelist" Jacob Freeman confirmed that DLSS 5 takes a 2D frame plus motion vectors as input. Freeman told a YouTuber that "DLSS 5 is trained end to end to understand complex scene semantics... all by analysing a single frame." This directly contradicts Huang's assertion that DLSS 5 operates at the geometry level with direct developer control over 3D data.
Developer trust erodes
The technical contradiction is compounded by a broader trust deficit among game creators. According to Insider Gaming, developers at Capcom and Ubisoft learned about Nvidia's use of their games in DLSS 5 demos at the same time as the public, rather than through advance notification from Nvidia.
This lack of communication proved particularly problematic for Capcom, which has historically positioned itself as anti-AI in game development. Some staff reportedly worry the DLSS 5 announcement could push the publisher toward accepting generative AI tools—a significant cultural shift.
The initial demonstration drew widespread criticism from developers themselves. Kotaku reports that one game developer said seeing DLSS 5 made them feel "there is no future for me" in the industry. Steve Karolewics, a rendering engineer at Respawn, described DLSS 5 as "an overbearing contrast, sharpness, and airbrush filter" that replaces original artistic intent. Former Rockstar developer Mike York called the technology "a complete AI re-render" that changes what players actually see on screen.
The core grievance isn't about AI as a concept. It's about artistic autonomy. When DLSS 5 automatically reshapes character models, adjusts lighting beyond what developers intended, or softens geometric detail, it exercises creative authority over work that developers spent months or years refining.
Where the argument stands
Huang's Lex Fridman interview addressed this directly. He emphasised that DLSS 5 is designed to be used at the development stage, not as a post-processing filter applied after games ship. He pointed to the technology's customisation options, including intensity sliders and colour grading controls, that allow developers to shape DLSS 5's output.
But this explanation creates more questions than it answers. If the technology relies on 2D input—as Freeman's comments suggest—then intensity controls and colour grading are fine-tuning tools applied after the AI has already made its core inferential choices about what a scene should look like. The geometry and artistic intent are already modified by the model.
Nvidia faces a credibility problem that no amount of diplomatic rhetoric can fix. The company showed off DLSS 5 applied retroactively to games like Resident Evil Requiem and Starfield without developer involvement, producing results many in the industry consider visually jarring. Huang can claim developer control is paramount, but the initial presentation suggested otherwise.
The company could have avoided much of this backlash with a more careful announcement: deeper technical transparency from the start, notification to partner studios before public reveal, and demos of the technology integrated properly into games built with DLSS 5 in mind, not retrofitted to existing titles.
Whether DLSS 5 becomes industry standard or a cautionary example of technology arriving before cultural acceptance will depend on how studios choose to implement it. But Huang's shift toward empathy, while welcome, doesn't address the underlying technical inconsistencies or the legitimate concern that generative AI models trained on internet data may carry aesthetic assumptions that override the intentional choices of human creators. That's a question tone-deaf responses alone cannot resolve.